Protocol for Stage Illumination, for use over IP networks
draft-mertl-psi-00

Abstract

This memo describes a protocol that builds upon UDP/IP to transport illumination and control data for stage, architectural and other illumination requirements.

It may be understood as a spiritual successor of the traditional DMX (Digital MultipleX) specification series, removing it's limitations and adding flexibility and usability enhancements like auto-discovery and metadata, among other useful features.

Status of this Memo

This Internet-Draft is submitted in full conformance with the provisions of BCP 78 and BCP 79.

Internet-Drafts are working documents of the Internet Engineering Task Force (IETF). Note that other groups may also distribute working documents as Internet-Drafts. The list of current Internet-Drafts is at http://datatracker.ietf.org/drafts/current/.

Internet-Drafts are draft documents valid for a maximum of six months and may be updated, replaced, or obsoleted by other documents at any time. It is inappropriate to use Internet-Drafts as reference material or to cite them other than as “work in progress.”

This Internet-Draft will expire on November 24, 2010.

Copyright Notice

Copyright (c) 2010 IETF Trust and the persons identified as the document authors. All rights reserved.

This document is subject to BCP 78 and the IETF Trust's Legal Provisions Relating to IETF Documents (http://trustee.ietf.org/license-info) in effect on the date of publication of this document. Please review these documents carefully, as they describe your rights and restrictions with respect to this document. Code Components extracted from this document must include Simplified BSD License text as described in Section 4.e of the Trust Legal Provisions and are provided without warranty as described in the Simplified BSD License.

Table of Contents

1.  Introduction
2.  Description
    2.1.  Groups
    2.2.  Clusters
    2.3.  Current Values
    2.4.  Safe Values
    2.5.  Message length
3.  Design
    3.1.  System-independent representation (transfer syntax)
    3.2.  Versioning
    3.3.  Identifier for nodes: IN (Identification Number)
    3.4.  Message types
    3.5.  Headers
        3.5.1.  Message header
        3.5.2.  Node header
        3.5.3.  Sentence Header
    3.6.  Word Types
    3.7.  Constants
        3.7.1.  Conventions
        3.7.2.  Protocol version
        3.7.3.  Bases for constants
        3.7.4.  Message types
        3.7.5.  Node Options
        3.7.6.  Sentence Options
        3.7.7.  Sentence types
        3.7.8.  Detailed Reactor Status
        3.7.9.  Channel status
        3.7.10.  Reactor types
        3.7.11.  Channel types
    3.8.  Dynamics
        3.8.1.  Discovery Phase
        3.8.2.  Initialization Phase
        3.8.3.  Normal Operation
4.  Discussion
5.  Refresh rate
6.  Routing Considerations
7.  Security Considerations
8.  IANA Considerations
9.  References
    9.1.  Normative References
    9.2.  Informative References
§  Author's Address

1.  Introduction

This document contains the specification of a protocol for stage illumination. The intent of this protocol is to be as useful and easy to use as [DMX512‑A] (ESTA, ESTA., “Entertainment Technology - USITT DMX512-A - Asynchronous Serial Digital Data Transmission Standard for Controlling Lighting Equipment and Accessories,” 2004.) while doing away with the severe limitations of DMX in terms of flexibility, structure and speed, while also adding several convenience and management features. The most important ones being:

• virtually unlimited number of nodes
• automated (re-)discovery of all nodes at any time
• larger and distinct data fields, including metadata
• fully bidirectional communication, including reception of data by the PSI Master
• ability to conglomerate individual PSI Reactors into Groups for improved transfer efficiency, as well as simultaneous unicast addressing as individual PSI Reactors, in any combination
• ability to use a very wide range of readily available, even stock, components as transport medium, leading to immediate adaptability to any environment
• may be run alongside other traffic, enabling the use of DHCP, or HTTP configuration in PSI Reactors, over the same network infrastructure
• while not recommended, PSI may be run across a preexisting, general network infrastructure including domestic networks without interfering (within limits of bandwidth requirements)

Comments are solicited and should be addressed to the author.


The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", "SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY", and "OPTIONAL" in this document are to be interpreted as described in [RFC2119] (Bradner, S., “Key words for use in RFCs to Indicate Requirement Levels,” March 1997.).

2.  Description

The PSI operates on top of existing protocols, namely UDP/IP. It is bidirectional and geared towards the typical environment found in stage lighting systems. The theoretical maximum number of nodes is given by the IN (Identification Number) that is eight bytes in size (see Section 3.3 (Identifier for nodes: IN (Identification Number))). In practice, bandwidth and memory limits are expected to set the actual limit for any given installation. A small number, usually exactly one, of controlling nodes, referred to as "PSI Masters", query, prepare, and send values to a potentially vast number of other nodes, referred to as "Reactors". Even though a PSI Master may have multiple network interfaces for various reasons like limiting bandwidth requirements, the entirety of Reactors connected to all it's interfaces is normally available as a whole, and is referred to as "Nexus". A PSI Master implementation may however choose to split the Nexus in any way deemed useful. In addition to data, information about all Reactors is gathered by the PSI Master, and can be presented to the user to aid in management and other tasks. Especially useful for this are the user-defined string and the user-defined numbers, because they may be used to indicate physical whereabouts and other distinguishing properties of any Reactor, in addition to a similar set of data that is vendor-specific.

Because the PSI does not use the underlying networks in unconventional ways, it will coexist peacefully, within bandwidth limits, with any other protocol running over the same networks. This enables, for example, the use of HTTP for configuration interfaces and DHCP for IP address assignment. Because a Nexus does not change behavior much during normal operation, even streaming media may be run alongside PSI.

To achieve a significant degree of flexibility and safety, it does not suffice to assign a channel number through the index of the data inside the data stream, like it is done in DMX ([DMX512‑A] (ESTA, ESTA., “Entertainment Technology - USITT DMX512-A - Asynchronous Serial Digital Data Transmission Standard for Controlling Lighting Equipment and Accessories,” 2004.)) , nor on a per-Reactor basis. While that method has the advantage of lower overhead, it has several disadvantages. One is that each message needs to contain every single channel, at least up to the one actually intended for changing. If channel 512 is supposed to be changed, all 511 previous channels must be sent as well, even if they are not even assigned to any device.

Additionally, a damaged, missing or misplaced value can not be detected, so that transmission errors may easily lead to incorrect settings in the receiving devices. Explicit channel numbers, on the other hand, allow for selectively sending data for arbitrary channels and in any order and make sure that a datum received truly is intended for a particular channel, instead of possibly being the result of a transmission error.
Additionally, UDP's checksumming is automatically used, and any intrinsic error detection and correction methods offered by the network are also taken advantage of. For example, PSI's expected primary deployment platform, Ethernet, uses a 32-bit checksum that provides a great deal of reliability to received data. The addition of a 16-bit checksum in the extended options of DMX proves the necessity of such reliability. While DMX needs to cope with legacy devices and thus can only make this checksum optional, it always has been mandatory in Ethernet.
In addition to that, PSI binds each channel to a specific Reactor, which is not possible with DMX. This way, the user can at any given time check and control, through the PSI Master, which Reactor contains any given channel, while in DMX this can only be achieved by physically examining every single device. The PSI Master allows the user to spot and resolve conflicts without moving, and without changing settings in Reactors. This is especially convenient because settings like these otherwise necessarily make use of vendor-specific interfaces, requiring the user to first study the device documentation before being able to make the necessary changes. The PSI puts this configuration into the PSI Master, where the user actually is located.

If a device needs to be replaced, the use of DMX requires that the new device is configured to mimic the replaced one. This will lead to problems if the replaced device supported functions the new one doesn't support, like assignment of non contiguous channel numbers. In such a case the entire controlling sequence, and / or other, completely unrelated devices, would need to be changed resp. reconfigured, resulting in a lengthy and error-prone management effort.
Through explicit assignment of channel numbers, paired with binding of channels to their containing devices, a change like that can be done easily: it only requires the physical replacement of the device in question and reassignment of that device's channels to the new device. If the PSI Master provides an abstract channel plane that is above the physical devices, then the logical arrangement created on top of the plane never needs to be touched. This means that any logical arrangement may be used, unchanged, on any Nexus, as soon as an appropriate abstract plane has been created through the PSI Master.

While DMX controllers may also support such an abstract plane (called "soft patching"), they can only map logical channels to DMX channels, without real relation to physical devices present. This therefore still leaves the problem of non contiguous channel ranges possibly forcing reconfiguration of many unrelated devices, on top of the need to change the soft patching. So, with DMX the user needs to maintain an updated list of all devices present, plus their configuration, to be able to configure new devices and to manage the soft patching, whereas the PSI Master automatically maintains this list. The list available from a PSI Master therefore is instantly available and, more importantly, always up to date. PSI creates this list on each restart from the devices actually present, and is able to flag removed devices and list newly added devices, without changing the current configuration. Newly (re-)appearing devices can also be dynamically added to the appropriate list, without changing the currently loaded patching.

The meta datatypes defined by PSI allow for a defined and therefore universal way to gather detailed information about the devices present on any given Nexus. These include vendor and device type, channel counts, and data types as well as user assigned identifiers and comments and detailed status information, allowing for a continuous overview of the entire Nexus.

Another advantage is the availability of larger data types, which are particularly useful for devices like scanners, moving heads and anything requiring finer resolution than the brightness of a lamp does. The development of (non-standard) 16-bit DMX by interpreting two consecutive 8-bit values as single 16-bit value shows that such larger data types are needed in many cases. Should even larger or different data be required, PSI provides additional data types in a well-defined and universal way, allowing for easy use of, for example, 64-bit values.

If a channel cannot handle the full range of possible values offered by the Word type required, it informs the PSI Master of it's value boundaries so it's limitations can be honored. A channel MUST tolerate reception of values outside these boundaries, and in response switch to Safe Values and set the appropriate Reactor Status.

Additionally, devices requiring unconventional data types may be employed without reinterpreting stock data types, by using the variable length data types defined by PSI. These may be used for images or similar data, but their actual interpretation completely depends on the device using them.

Each message sent to the PSI Master also contains a crude status indicator, allowing for quick notification about unusual circumstances, possibly automatically triggering more detailed inquiries and / or other measures by the PSI Master. This is especially useful in large installations, because such a Nexus cannot normally be satisfactorily presented as a whole. Since the PSI Master always possesses complete and up to date information, it is able to actively notify the user of any irregularities in all cases.

2.1.  Groups

Groups are used to efficiently handle installations that are composed of many Reactors with few channels each. Sending data to each of them in a separate message may unduly impact the available bandwidth, so it may make sense to handle a number of such Reactors in a single broadcast or multicast message (see Section 3.4 (Message types)). To do so, an identifier is assigned to all Reactors that the new Group shall contain. This identifier may then used whenever a message applies to more than one member of the Group. Use of this identifier allows all Reactors that are not part of the specific Group to quickly notice that fact and thus discard the message without having to process it any further. A message directed at members of a Group contains the usual identifiers for the individual Reactors and may therefore convey totally different data to each member.

2.2.  Clusters

Clusters are Groups applying the same data to all members (indicated by the flag PSI_NO_FM_CLUSTER being set). From PSI's point of view, a Cluster is exactly the same as a Group. However, from the user's point of view, both are distinct. While Groups serve to more efficiently utilize bandwidth in case of large numbers of Reactors with few channels, Clusters serve to efficiently apply the same data values to multiple Reactors, so these Reactors may have a large number of channels but need to be reasonable similar to each other, especially in terms of channel configuration. Therefore, a PSI Master that implements Clusters should distinguish Clusters and Groups, and apply sanity checks whenever the user adds members to a Cluster. Specifically, Clusters composed of Reactors using different channel configurations or data types should require explicit user confirmation. It may allow this regardless, because the Cluster flag MAY be used to convey general instructions (like Node Options), or to set only a limited subset of channels that may be applicable to all Reactors in the Cluster. For example, all Reactors in a Cluster may have a channel that is mapped at number 10 and uses 32 bit data. So even though the remaining channels of the clustered Reactors may differ, this single channel may be assigned values using a Cluster message. The PSI Master SHOULD check if there are any matching channels in all clustered Reactors when a new one is added to the Cluster, and inform the user of the result. It MAY offer the user a filter to display only the matching channels.

Also, the CLUSTER flag need not be used in all, even most, messages sent to a Reactor. It is perfectly acceptable to, for example, use this flag only to assign values to a single channel out of many, and to do so only occasionally. Of course, the more this flag is used, the higher transmission efficiency will be.

2.3.  Current Values

The Current Values are what is being sent by the PSI Master to a Reactor. They are the result of calculations and / or inputs of any sort. Because they can be any value of the possible value range, they are not suited for continued use in case of loss of control. Instead, Safe Values must be used in that case.

2.4.  Safe Values

The Safe Values are specific to each channel. They are designed to be used whenever a channel receives no updates from the PSI Master, and can be used continuously without posing danger for the device or audience.

2.5.  Message length

Because fragmenting of messages leads to greater impact of packet loss, it should normally be avoided. In order to keep IP from fragmenting messages, the default maximum message length should be 1400, since 1500 is the normal MTU for Ethernet, allowing some room for different configurations.

A Reactor implementation MAY allow the user to change this maximum, in response to the actual network performance, but it SHOULD NOT be less than 1400. In fact, the maximum message length of Reactors may be significantly larger by default, because the PSI Master is free to stay well below the maximum, and therefore can adapt, possibly automatically, to network loss rate. On the other hand, the PSI Master's maximum message length may change, and an information message is sent to the Reactors, during normal operation, so that the Reactors do not need to be manually reconfigured to stay below a certain size.

An embedded platform therefore is free to choose it's absolute maximum message length, so that it can use static buffers, but advertise less than that if deemed reasonable.

3.  Design

This section describes the construction of messages sent over the media.

3.1.  System-independent representation (transfer syntax)

Experience using large and heterogeneous data networks has shown that reliable and easily adjustable communication between devices requires a well-defined standard. Especially, standardizing the endianness does not suffice. Instead, a system- and language-independent data representation is required. A number of such representations have been examined for use as basis for PSI, and unaligned CDR was chosen (see Section 4 (Discussion) for details).

CDR does not specify the endianness of bit fields, so they are defined by PSI to match the endianness of the remainder of the message.

3.2.  Versioning

Even though multiple versions are not planned, and should not be created without proper deliberation, every PSI message, including discovery messages, contains a version field. It MUST be checked by every receiver against it's list of supported versions to determine how the remainder of the message needs to be interpreted. If a Reactor supports more than one version, it SHOULD respond using the version used by the PSI Master. If the Reactor does not support the version used by the PSI Master, it discards its messages. For every message received in an incompatible version, the Reactor sends a Version Mismatch message (see Section 3.7.4 (Message types)) to the sender. The PSI Master must flag any version differences for the user to see, and do so prominently for incompatibilities.
Likewise, a Reactor that doesn't use it's preferred version should, if it provides other indicators to the user, indicate a warning, and an error if it does not have a compatible version to use. A node may support any number of different versions, and allow the user to selectively disable their use, and to specify which is the preferred version.

In order for a node running any protocol version to be able to detect and report version incompatibilities as well as to send Version Mismatch messages to the sender, the message header needs to maintain the structure and contents described in this memo for all versions created. It therefore MUST be modified only in ways that keep the structure in place, like by appending to the basic format. The contents of the basic structure must be made as meaningful as possible.

3.3.  Identifier for nodes: IN (Identification Number)

Each node needs to be uniquely identifiable and thusly referable within the entire Nexus, for example by a number. The uniqueness of this identifier is very important to guarantee the correct functioning of the installation. To facilitate this without requiring lengthy and error-prone configuration of each Reactor, it needs to even be globally unique. Since PSI is geared towards Ethernet, making use of the MAC-address of the first NIC found in the node makes sense, as it is globally unique already. Regardless, the user may assign the INs in any way convenient as long as they are unique within the Nexus.
Despite the MAC address of Ethernet and most other media types being six bytes in length, this field is defined to be 8 bytes in length to accommodate the 8 byte MAC format (EUI-64) and to facilitate other uses, like multiple personalities (PSI Master and Reactor). The MAC byte sequence is filled in in canonical format, with the OUI and the remainder of the MAC separated by as many filler bytes with the value 0xFF (and, if necessary, bits with value 1), that both the OUI and the remainder of the MAC address occupy the outmost bytes. This conversion therefore is compatible with the IEEE's suggested practice, albeit making no distinction between MAC-48 and EUI-48.
Since this number has no corresponding data type, it is defined as sequence of 8 bytes. This is no problem in this case, because sequences of bytes are stored the same on all architectures, meaning that no endian issues arise from this (see Section 3.1 (System-independent representation (transfer syntax)) for details). Through the IN, identification of nodes does not depend on IP address assignment, so that may be done through DHCP.
In the event that a single node runs more than one personality (PSI Master / Reactor) at the same time, each of them MUST use a different, unique, IN. In a Nexus using the 6 byte MAC format exclusively, this can be accomplished easily by using the two filler bytes as an index. Alternatively, and especially when using the 8 byte MAC format, the user may assign artificial INs in any way convenient, provided they are unique within the Nexus.

3.4.  Message types

PSI uses a number of basic message types that are further split by direction (to or from the PSI Master). This allows for quick checking whether the received message can be intended for the receiving node at all. For example, a *_FM_* message can not be intended for a PSI Master, not even by other PSI Masters. Additionally, the meaning of some message types differs depending on direction. Any message may only contain those elements (like sentence types, Node Options, etc.), that apply to it's direction, because some differ greatly, or even conflict, depending on direction. Therefore it is necessary to check for the proper direction before using data from a received message.

The various existing and allowed message types are as follows:

• Unicast message
  
  This is the basic message type and is usable for all nodes. It is sent directly to a single node and therefore creates little processing overhead for any other nodes. Additionally, the entire content of the message is meaningful to the receiving node, so no irrelevant portions require processing. Therefore, messages of this type must only contain information for a single node. However, they may be spread over multiple Node Sections. Use of the flag NODE_FINISHED is optional and does not influence the transmission efficiency. The unicast message uses no GID so this field is omitted in the message header. It is possible to send this message type as multicast or broadcast, using multiple Node Sections with different TINs. However, this method creates processing overhead for all (matching) nodes and should thus be avoided in favor of the more appropriate message types using a GID.
  
  The allowed contents differ depending on direction for this message type, see Constants (Section 6 (Routing Considerations)).
  
  
• Multicast message
  
  By using this message type the multicast capability of the transport layer may be utilized to send, relatively targeted, messages to a Group. In theory, this could be a middle ground between unicast and broadcast, though in practice this may require careful address and GID assignment. If used, it can reduce the impact of Groups on CPU usage in nodes; there is otherwise no difference between this and a broadcast message.
  
  The allowed contents differ depending on direction for this message type, see Constants (Section 6 (Routing Considerations)).
  
  
• Broadcast message
  
  This message type is intended for use in conjunction with Groups, so that the bandwidth can be used more efficiently in installations with many Reactors of only few channels each. A message of this type therefore can contain multiple Node Sections encapsulating contents for different nodes. As in unicast messages, it is allowed to send multiple Node Sections for the same node in a single message. To increase efficiency, the last node header preceding data for any specific node SHOULD have the flag NODE_FINISHED set, so the node can discard the remainder of the message immediately. It is recommended that all Node Sections applying to any specific node be consecutive, since that further reduces overhead for all but the last node addressed in the message.
  
  The allowed contents differ depending on direction for this message type, see Constants (Section 6 (Routing Considerations)).
  
  
• Discovery message
  
  This is a special message type that must not contain additional data. It's use is restricted to, depending on direction, inform PSI Masters of newly appearing Reactors, or to instruct all Reactors present to send discovery messages, allowing a new or restarted PSI Master to build and maintain an inventory of the Nexus (See Section 3.8 (Dynamics) for details). Another use is to directly contact a PSI Master in case it ceased sending messages to a Reactor without apparent reason. This way, the Reactor can initiate being (re-)added to the PSI Master's inventory if, for example, the PSI Master has been restarted but no global discovery request has been made, or in case the Reactor has missed it (it may have been disconnected from the network at that time).
  A Reactor receiving a discovery message SHOULD switch to Safe Values; channels using Safety Sequence MUST do so, cease operation, ensure safety if at all possible and also MUST discard any progress through their Safety Sequences.
  This message may therefore be sent, depending on the sending node's requirements, as uni-, multi, or broadcast, and is therefore not further subdivided.
  
  
• Version Mismatch message
  
  This is a special message type that must not contain additional data. It's use is to inform the receiver that the sender does not support the protocol version that was used to contact the sender. The version field contains a version supported by the sender. If the sender supports multiple versions, then the first Version Mismatch message contains it's preferred, default version. If the sender receives another incompatible message, then it sends the next version that it supports, until it starts over at the preferred version.
  The receiver of such a message will check if it supports the version indicated, and if it does, then it will establish communication using that version. If it does not, it will collect all versions supported by the sender by sending messages to the sender. The list is complete if the sequence has repeated at least once, possibly more often to allow for duplication, loss and out of order delivery. The receiver may then check this list against it's own support list and pick the version that is nearest to the beginning of both node's lists, or by any other criteria, if multiple matching versions exist. This way, it is ensured that communication is established eventually, and also that, if more than one possible match exists, the version picked is always the same.
  
  The reason this is done this way is to have as few message types and contents that must be maintained compatibly across all versions as possible. Using Node Specification to list all supported versions in order of preference, for example, would not only fix that Word type, but also force the entire structure of at least the unicast message to be essentially fixed.

3.5.  Headers

PSI uses three headers that represent different layers of the protocol and are related in a strictly hierarchical manner.

3.5.1.  Message header

The message header represents the highest layer of the PSI. Every message must have exactly one message header, as it contains all necessary information like protocol version, source and the type of the message. Additionally, it allows for checking the completeness of the received message in terms of length. The method of specifying the source of the message instead of it's destination is uncommon and stems from the requirement that the protocol must be able to send information for different Reactors in a single message, in order to reduce the message overhead for Reactors with only minimal data requirements (like those with only a couple of channels). Otherwise messages, and therefore Ethernet frames, would need to be sent for even a single channel. This way, messages for several Reactors may be sent as directed broadcast or multicast, significantly reducing the number of messages that need to be sent. The downside of this is that broadcasts place a processing burden on all connected nodes, since they will need to interpret the message until they can conclude that it does not contain information for them. The increasing processing power of embedded equipment likely will reduce the impact of this, but likewise will the increasing bandwidth of even commodity networks reduce the impact of sending many, tiny messages. In the end, it will depend on the particular circumstances and thus should be kept in mind when setting up the PSI Master's configuration.

To further alleviate this problem, PSI uses an additional data field in the message header of broadcast and multicast messages, the Group IDentification (GID). Through use of the GID, not every Reactor needs to check each received broadcast / multicast message for relevant information. Instead, it may abort the check if the GID in the received header does not match any of the GIDs assigned to the Reactor, if there are any.
Only messages of type PSI_MT_xx_MULTICAST and PSI_MT_xx_BROADCAST contain a GID-field.

A PSI Master implementation MAY allow for such conglomeration, be it manual and / or automatic. In any case, such conglomeration may be changed at any time, including Reactors being assigned different GIDs. If a PSI Master allows for conglomeration, it MUST also provide means of disabling it, being part of the unicast-only mode of operation (see Routing Considerations Section 6 (Routing Considerations)).

Over the wire format:

 0                   1                   2                   3
 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
|    Version    |     Type      |         Message Length        |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| : . Message Length (ctd.) . : |              SIN              |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
|                          SIN (ctd.)                           |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
|          SIN (ctd.)           | : . : . : . :GID: . : . : . : |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| : . : . :GID (ctd.) : . : . : |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+

Legend:
: . XYZ . : = bytes that only exist when certain flags are set

The PSI message header.

Detailed description of the header fields:

Version: 8 bit

This field contains the version of the sending PSI stack. This is used to determine if the message can be interpreted, and if so, how. See Section 3.2 (Versioning) for details.

Message type: 8 bit

This field specifies the type of the message. The basic types are subdivided into two subtypes each, that define the message direction (from the PSI Master or to the PSI Master); see Section 3.4 (Message types) for a list and descriptions. These subtypes differ in what their contents may be, and also allow testing for logical correctness. For example, a message sent to the PSI Master cannot be intended for it if it is of subtype "from PSI Master". This is especially important in the case of broadcast messages, because it allows faster weeding out of irrelevant messages.

Additionally, the endianness and the size of all length fields in the message is defined through bits 1 and 2, respectively, of this field:

For all message types, bit 1 of the "message type" field indicates the endianness of the entire message (including bit fields): if bit 1 is set, then the message is in big endian format, otherwise the message is in little endian format. The receiver therefore only needs to convert if the sender uses the other format. An implementation MAY allow the user to choose which endian type to use for sending, regardless of the architecture's natural format. For example, if most Reactors use one format while the PSI Master uses the other one, the PSI Master could be configured to use the predominant format for sending, instead of it's native format.

For all message types, bit 2 of the "message type" field indicates the size of all length fields in the entire message: if bit 2 is set, then the length fields are 32 bits in size, otherwise their size is 16 bit.
This is done because in nearly all cases, the maximum message size will be around one Ethernet MTU, or at least well below the 65536 bytes possible using 16 bits, so for each length field, two bytes less bandwidth are required. Since the message components (Node Sections and Sentences) can only be smaller than the total message length, their length fields will never need to be larger than the one of the message header. In the uncommon case when truly large chunks of data need to be sent in one message, then 32 bits may be used for all length fields. In that case, the additional bytes for the length fields will not impact performance due to the large amount of data to be sent.

Thus, including the message direction (bit 0, see Section 3.7.3 (Bases for constants)), the most significant 3 bits are used as flags.

Message length: 16 / 32 bit

This field contains the length of the message in bytes, including the message header with the length field itself. The receiver compares it to the number of bytes received to see if the message is complete.

SIN: 8 byte

This is the Source Identification Number and contains the sender's IN (see Section 3.3 (Identifier for nodes: IN (Identification Number))). It informs the receiver of the source of the message. It is used, depending on the message type, for checking (for example, if the sender actually is allowed / supposed to send a particular message type or contents to the receiver) or, especially in case of Discovery messages, to maintain the list of PSI Masters in the Reactors, and the list of Reactors in the PSI Master(s).

GID: 32 bit

The Group ID is specific to broadcast- and multicast message types. It narrows the scope of the message to only a small subset of the receiving Reactors. By simply comparing the GID to the list of GIDs assigned to itself, a Reactor can decide if it needs to interpret the remainder of the message at all. If a Reactor has no GIDs assigned, it may ignore multicast- and broadcast messages if they are identified as such by their type, since they cannot contain information for that Reactor.

Assignment of GIDs is done by the PSI Master and may be changed during normal operation, for example when Reactors vanish or new Reactors appear. The field's size of 32 bit allows for about 4.2 billion unique Groups, which suffices for even large systems.

3.5.2.  Node header

The node header is used to identify the target Node and marks the beginning of a Node Section. Each message, except for Discovery messages, may contain any number of Node Sections that, depending on the message type, may apply to the same node and / or to different nodes. To facilitate easy jumping across irrelevant (to any particular Node) sections, and to allow for checking completeness and correctness, each node header contains the length of the entire Node Section. It also contains the Target Identification Number ("TIN") as well as a field for options that govern the intended use of the data grouped under this header, or represent simple instructions for the entire node. Due to the last part, a node header may be solitary, meaning it does not need to be followed by any Sentence Headers (Section 3.5.3 (Sentence Header)).

Over the wire format:

 0                   1                   2                   3
 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
|                         Node Options                          |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
|         Section Length        + : . Section Length (ctd.) . : |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
|                              TIN                              |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
|                          TIN (ctd.)                           |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+

Legend:
: . XYZ . : = bytes that only exist when certain flags are set

The PSI node header.

Detailed description of the header fields:

Node Options: 32 bit

This is a bit field that contains instructions for the use of the data contained in the current Node Section, as well as self-contained instructions and information. This usually are queries that apply to the entire node, or the flags NODE_FINISHED and REACTOR_ACCEPTED. A description of all Node Options is found in Section 3.7.5 (Node Options).

Section Length: 16 / 32 bit

Similar to the length field of the message header (Section 3.5.1 (Message header)), this field contains the length of the Node Section, including the node header and the length field itself. It allows for checking for completeness and correctness, and easy jumping across parts not interesting to the particular node. Since the next Node Section starts, relative to the start of the length field, at an offset of exactly the amount of bytes named in the field, this weeding out can be done relatively easily.

TIN: 8 byte

This is the Target Identification Number and contains the target's IN (Section 3.3 (Identifier for nodes: IN (Identification Number))). It defines the intended receiver for the current Node Section. If the TIN in any given Node Section does not match the receiver's IN, it must not be interpreted or otherwise used. In this case, a simple jump across the entire Node Section makes sense, using the length field, to not waste CPU time. An implementation MAY also choose to process the Node Section the normal way and just discard the extracted data afterwards, for example if a thorough consistency check is desired. The method chosen has no effect on the data transferred or other nodes.

If the CLUSTER flag is set, the TIN field MUST NOT be present.

3.5.3.  Sentence Header

The Sentence Header contains information that applies to the actual data transferred, in order to precisely identify them. It controls which type of Words make up the subordinate sentence. Any given sentence must contain only Words of the same type, but a Node Section may contain any number (including zero) of sentences of any type.

Over the wire format:

 0                   1                   2                   3
 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| Sentence Type |               Sentence Options                |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
|Se. Opt. (ctd.)|        Sentence Length        + :S. L.(ctd.): |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| :S. L.(ctd.): |
+-+-+-+-+-+-+-+-+

Legend:
: . XYZ . : = bytes that only exist when certain flags are set

The PSI sentence header.

Detailed description of the header fields:

Sentence Type: 8 bit

The sentence type specifies the type of the Words making up the current sentence. It is the same for all Words of any given sentence and can thus be used, in combination with the sentence Length field, to compute the Word count. This saves an additional field for the Word count. See Section 3.6 (Word Types) for a description of all Word types.

Sentence Options: 32 bit

This field is, like the Node Options field, a bit field, that specifies the data contained in the current sentence as well giving instructions to their use. It is especially used to govern the use of generic Word types like Node Specification and Channel Specification. Additionally, it may contain requests or instructions that are to be applied to the data contained in the sentence. This way, the PSI Master can use a single sentence of type Channel Number to request all information about the channels specified, as well as switch the channels between Safe Values and Current Values. By using Words of one of the Data Word types, new values may be assigned to the specified channels at the same time.
In messages to the PSI Master, this field merely specifies the use of the generic Word types.

Sentence Length: 16 / 32 bit

Like the length fields in the message- and node headers, this field contains the number of bytes used by the current sentence, including the Sentence Header and it's length field. Contrary to the lengths of Node Sections and Messages, it is not meant to facilitate jumping across the sentence, since the relevance of any particular sentence has already been established by the previous examination of the preceding node header. Instead, this field describes the number of Words contained in the sentence. To be consistent with the other length fields and to allow for easy checking of consistency, it contains the total byte count.
If the length of the current Word type, which is given in the sentence type, is fixed, then the number of Words contained can be derived through simple division, allowing further check for consistency of Word type and sentence length. If the length of the current Word type is variable, like for example the Reactor Status, the sentence may contain only exactly one Word, with the sentence length representing the actual length. This restriction stems from the deliberation that Words of variable length will be used only occasionally, with their length possibly becoming comparatively large. Therefore, the overhead of a complete Sentence Header per Word is acceptable in this case. Since the majority of Words transferred is of fixed length however, the introduction of an additional length field for each Word would cause comparatively large overhead, while it's usefulness would be restricted to comparatively few and seldom used Words and thus marginal. Especially with small Words, like channel numbers, the overhead would be 100% of the actual data and therefore not justified.

3.6.  Word Types

To further explain the structure of PSI messages, this section lists and describes the types of Words used. The sequence described is the sequence as seen on the network. The ordering of bytes and other issues are governed by the transfer syntax (Section 3.1 (System-independent representation (transfer syntax))) and therefore not explicated here. Channel number sizes are given by the appropriate Sentence Option. The Word type used, in conjunction with the data boundaries, defines the resolution to use, so the PSI Master knows which range of values is allowed for the channel and can scale the values appropriately. Values within the Word type but outside the defined boundaries MUST make the channel switch to Safe Values and setting of the appropriate Reactor Status.

While a PSI Master MUST implement all of these types except where mentioned otherwise, a Reactor only needs to implement the types of PSI_Word_Data that it actually uses.

Boolean Data Word: PSI_Word_DataB

This data Word consists of the channel number and the actual, boolean, datum.
These two syllables are sent in the sequence "channel number, datum".

Signed 8 bit Data Word: PSI_Word_DataS8

This data Word consists of the channel number and the actual, signed, datum, with a width of 8 bits.
These two syllables are sent in the sequence "channel number, datum".

Unsigned 8 bit Data Word: PSI_Word_DataU8

This data Word consists of the channel number and the actual, unsigned, datum, with a width of 8 bits.
These two syllables are sent in the sequence "channel number, datum".

Signed 16 bit Data Word: PSI_Word_DataS16

This data Word consists of the channel number and the actual, signed, datum, with a width of 16 bits.
These two syllables are sent in the sequence "channel number, datum".

Unsigned 16 bit Data Word: PSI_Word_DataU16

This data Word consists of the channel number and the actual, unsigned, datum, with a width of 16 bits.
These two syllables are sent in the sequence "channel number, datum".

Signed 32 bit Data Word: PSI_Word_DataS32

This data Word consists of the channel number and the actual, signed, datum, with a width of 32 bits.
These two syllables are sent in the sequence "channel number, datum".

Unsigned 32 bit Data Word: PSI_Word_DataU32

This data Word consists of the channel number and the actual, unsigned, datum, with a width of 32 bits.
These two syllables are sent in the sequence "channel number, datum".

Signed 64 bit Data Word: PSI_Word_DataS64

This data Word supplies 64 bit for a signed datum.
The syllables are sent in the sequence: "channel number, datum".

Unsigned 64 bit Data Word: PSI_Word_DataU64

This data Word supplies 64 bit for an unsigned datum.
The syllables are sent in the sequence: "channel number, datum".

Floating point Data Word: PSI_Word_DataFP

This data Word supplies 32 bit for a floating point datum.
The syllables are sent in the sequence: "channel number, datum".

Double floating point Data Word: PSI_Word_DataDFP

This data Word supplies 64 bit for a floating point datum.
The syllables are sent in the sequence: "channel number, datum".

Quadruple floating point Data Word: PSI_Word_DataQFP

This data Word supplies 128 bit for a floating point datum.
The syllables are sent in the sequence: "channel number, datum".

Data Word of variable length: PSI_Word_Data_Opaque

This Word allows the transfer of unspecified data. It may be used for exotic systems that cannot be reasonably covered using the other Word types. It is necessary to note that no processing of the supplied data is performed, so that device-dependent issues are left for the communicating nodes to solve themselves. Merely the channel number is transferred as usual. Because the interpretation depends entirely on the receiver, these generic types are just asking for trouble and should be used only when truly necessary.

This is one of the Word types that have no fixed length and therefore may only occur once per sentence, requiring a sentence per Word. The reason is that the sentence length is used to convey the number of bytes making up the Word, which therefore represent the syllables of this type of sentence. See also Section 3.5.3 (Sentence Header).

It must be noted that the number of bytes is prepended implicitly by the transfer syntax, and thus in addition to the sentence length field. Therefore, the Sentence Length field needs to also contain the number of bytes for the implicit length field so that the message length is calculated correctly. By implicitly prepending the actual Word length, 32 bits of overhead are created. The receiver implicitly uses and removes the added length field. To facilitate custom structuring of the contents, the encoding used is sequence.
The syllables are sent in the sequence: "channel number, Word length (implicit), data bytes".

Data Word for device-specific data: PSI_Word_Data_Plain

This Word differs from PSI_Word_Data_Opaque only in that it does not contain a channel number.
The syllables are sent in the sequence: "Word length (implicit), data bytes".

Data Word for plain Channel Number: PSI_Word_Channel_Number

This Word contains only one syllable, representing a channel number. It is used in conjunction with Sentence Options to send instructions to and to query information about the channel(s) named.
The syllables are sent in the sequence: "channel number".

Data Word for Text: PSI_Word_Text

This is another Word without fixed length, and is thus only allowed to occur once per sentence because it's length is given by the Sentence Length. It is used, for example, for the user-assigned textual ID; it's syllables are single bytes that contain purely text data. It is converted appropriately if needed by any specific system architecture. In addition to the data and termination bytes, the transfer syntax prepends the actual number of bytes, creating an overhead of four bytes that also needs to be taken into account in the Sentence Length.
The syllables are sent in the sequence: "channel number, Word length (implicit), data bytes".

The encoding used is UTF-8 ([RFC3629] (Yergeau, F., “UTF-8, a transformation format of ISO 10646,” November 2003.)) (the sender MAY use plain ASCII), and it SHOULD be encoded using the string type.

Generic Data Word for numbers: PSI_Word_Generic_32

This Word also has no fixed length, and is given by the Sentence Length. It is used for transferring numbers like the user-assigned numeric IDs. All syllables are 32 bits wide and thus the Sentence Length always is a multiple of four bytes. Therefore it can be used to calculate the number of syllables and no additional field is needed to store the actual length.
The syllables are sent in the sequence: "datum, datum,...".

Data Word for Reactor Status: PSI_Word_Reactor_Status

This Word also has no fixed length, and is given by the Sentence Length. It is used for transferring detailed status information about a Reactor. All syllables are 8 bits wide and thus the Sentence Length can be used to calculate the number of syllables and no additional field is needed to store the actual length.
The syllables are sent in the sequence: "datum, datum,...".

Data Word for channel status: PSI_Word_Channel_Status

This Word contains two syllables: a channel number and the channel status, that is encoded as bit field of 16 bits.
The syllables are sent in the sequence: "channel number, channel status".

Data Word for Safety Sequence: PSI_Word_Safety_Sequence

This Word is used to convey the sequence required to change the value of a channel that requires extra safety. It has a variable length and thus must only occur once per sentence. It's component syllables are 32 bits in size and thus the Sentence Length can be used to calculate the number of syllables, without the need for an additional length field. The first syllable is the channel number; all following syllables are the numbers to send, in the same sequence as given in this Word, to activate the channel.
The syllables are sent in the sequence: "channel number, sequence component 0, sequence component 1,...".

Any such channel MUST NOT change it's value if the sequence is not received fully, out of order, or numbers not in the sequence are received. To change the value of such a channel, the sequence MUST be split into individual messages containing Words of this type, but only a single component of the sequence at a time. The value that the channel is supposed to switch to MUST be sent with the same message, in a separate sentence of the Word type as the channel's data type, with every component of the sequence. All messages that contain parts of the sequence MUST contain the same target value. If the target value in one of the messages does not match the previous, then, even if the sequence component itself matches the one expected, any sequence progress MUST be discarded and the value MUST NOT be changed.

This is done to ensure that no single erroneous packet can trigger a value change (unless the sequence length is 1, which is allowed but strongly discouraged). However, multiple channels, including channels using Safety Sequence, may be served in a single message.
Because any packet may get lost or received incorrectly, to ensure that the sequence has been advanced, the flag PSI_NO_FM_SENDACK may be used.

Additionally, a channel using Safety Sequence MUST NOT advance the sequence when it is using Safe Values, like after receiving PSI_SO_FM_GOSAFE or PSI_NO_FM_GOSAFE.

The actual length of the sequence is left to be assigned by the users of the system, governed to their safety requirements, so an implementation MUST NOT expect any specific length. A device MAY include a default sequence deemed reasonable by the vendor, but it MUST allow the user to change the sequence, including it's length. A length of 1 MUST NOT be used as default, and an implementation MUST warn the user about the safety implications of setting it to that length. A length of zero MUST make the channel ignore any value changes it may receive, effectively disabling it. A sequence MUST NOT contain any single number more than once, so that duplicated or reordered packets can not advance the sequence.
Whenever the sequence received does not match the required sequence, the receiver MUST discard all parts received already, and restart at the beginning. The numbers 0 and 0xFFFFFFFF (all bits set) MUST reset the receiver, and therefore cannot be used as part of a sequence. This is done so that any channel using Safety Sequence can safely be reset without knowing the sequence that has been assigned to it.

Note: it would be possible to add explicit numbering to each part of the sequence, but because the sequence length will always be small compared to the available number space, this seems to be unnecessary overhead, adding to the overhead already created by splitting.
A preliminary suggestion is for a channel implementation to have room for a length of at least 20 per channel. Any PSI Master implementing Safety Sequence is expected to allow for longer sequences.

Implementation of this type is completely optional.

Data Word for channel counts: PSI_Word_Channel_Count

This Word is composed of three syllables, each 64 bit in size, specifying the counts of the various channel types in a Reactor.
The syllables are sent in the sequence: "input channel count, in/out channel count, output channel count".

Generic Word for Node Specification: PSI_Word_Node_Specification

This Word only contains a single value of 32 bits. It's meaning is defined by the Sentence Options. This method was chosen to not have to define a large number of Words that differ in name only.
The syllables are sent in the sequence: "node specification".

Generic Word for Channel specification: PSI_Word_Channel_Specification

This Word contains two syllables: the channel number and a channel specification of size 32 bit. Like PSI_Word_Node_Specification, the actual meaning is given by the Sentence Options.
The syllables are sent in the sequence: "channel number, channel specification".

3.7.  Constants

This section lists and explains the constants used in PSI. Since only the values of the constants are transferred between nodes, an implementor may choose to use any convenient naming and assignment scheme.

3.7.1.  Conventions

The conventions used in this document are as follows:

• Naming
  
  All constants are prefixed with "PSI" to emphasize their belonging to the PSI, and to avoid collision and confusion with other constants. An underscore ("_") separates the specification of the general usage in the form "AB". Following another underscore is the direction in form "CD", if the constant is defined for more than one direction. A further underscore separates this from the general meaning of the constant, usually given by an acronym "E". Thus, constant names are constructed like this:
  
  PSI_AB_CD_E (directional constant), like PSI_ST_TM_WD32 PSI_AB_E (nondirectional constant), like PSI_CR_8
  
  Values for AB:
  
  PV: Protocol Version
  MT: Message Type
  NO: Node Option
  SO: Sentence Option
  ST: Sentence Type
  RT: Reactor Type
  RS: Reactor Status
  CT: Channel Type
  CS: Channel Status
  
  Values for CD:
  
  FM: From PSI Master
  TM: To PSI Master
  
  Constants that have different application (like single or conglomerate constants) are named in a way to show their meaning, and are prefixed with "PSI".
  
  
• Values
  
  Especially bit constants are given in the form "bit X", specifying the bit that is set, counting the MSB of the field as bit 0. Other constants, especially large ones, are given as hexadecimal, prefixed by the conventional "0x".

3.7.2.  Protocol version

PSI-32 Bit: PSI_PV_0

This is the protocol described here.
It is defined as 0.

3.7.3.  Bases for constants

To ease readability and structure, there are a few constants serving as base for others. The bases are chosen in a way to allow the constants of every area to start from 0 by adding the base as offset.

Type base for 'direction from PSI Master': PSI_TYPE_BASE_FM

This is the base for all constants that depend on the message direction, identifying data coming from the PSI Master.
It's value is 0.

Type base for 'direction to PSI Master': PSI_TYPE_BASE_TM

This is the base for all constants that depend on the message direction, identifying data sent to the PSI Master.
It's value is "bit 7", dividing the available number space by half.

Type base for 'message from the PSI Master': PSI_MT_BASE_FM

This is the base for all message types that are sent by the PSI Master.
It's value is PSI_TYPE_BASE_FM.

Type base for 'message from the PSI Master': PSI_MT_BASE_TM

This is the base for all message types that are sent to the PSI Master.
It's value is PSI_TYPE_BASE_TM.

Type base for sentences in messages coming from the PSI Master: PSI_ST_BASE_FM

This is the base for sentence types that are used in messages sent by the PSI Master.
It's value is PSI_TYPE_BASE_FM.

Type base for sentences in messages sent to the PSI Master: PSI_ST_BASE_TM

This is the base for sentence types that are used in messages sent to the PSI Master.
It's value is PSI_TYPE_BASE_TM.

3.7.4.  Message types

The following constants identify the message types as described in Section 3.4 (Message types). The direction of the message is identified by use of the appropriate constant. The meaning of a message may differ depending on it's direction, so this is not only important for plausibility checking.

As described in Section 3.5.1 (Message header), including the message direction (bit 0), the most significant 3 bits are used as flags.

Messages from the PSI Master:

Normal (unicast) message: PSI_MT_FM_NORMAL

This designates the most basic type of data message. The message is directed at a single Reactor only and therefore does not contain a field for a GID. It may, however, contain multiple Node Sections.
It's value is (PSI_MT_BASE_FM+0).

Multicast message: PSI_MT_FM_MULTICAST

This type designates a message intended for multicasting. Contrary to the unicast message, it contains a field for a GID.
It's value is (PSI_MT_BASE_FM+1).

Broadcast message: PSI_MT_FM_BROADCAST

This type designates a message intended for broadcasting. Like the multicast message, it contains a field for a GID.
It's value is (PSI_MT_BASE_FM+2).

Discovery message: PSI_MT_FM_DISCOVERY

This type designates a PSI Master to Reactor Discovery message. It does not contain information besides the message header itself.
It's value is (PSI_MT_BASE_FM+3).

Version mismatch message: PSI_MT_FM_VERSION_MISMATCH

This type designates a PSI Master to Reactor version mismatch message. It does not contain information besides the message header itself. The version field indicates the next supported version.
It's value is (PSI_MT_BASE_FM+4).

Messages to the PSI Master:

Normal (unicast) message: PSI_MT_TM_NORMAL

This message is directed at a single PSI Master. This is the usual case even in the presence of multiple PSI Masters.
It's value is (PSI_MT_BASE_TM+0).

Multicast message: PSI_MT_TM_MULTICAST

This type designates a message intended for multicasting. Contrary to the unicast message, it contains a field for a GID.
It's value is (PSI_MT_BASE_TM+1).

Broadcast message: PSI_MT_TM_BROADCAST

This type designates a message intended for broadcasting. Like the multicast message, it contains a field for a GID.
It's value is (PSI_MT_BASE_TM+2).

Discovery message: PSI_MT_TM_DISCOVERY

This type designates a Reactor to PSI Master Discovery message. It does not contain information besides the message header itself.
It's value is (PSI_MT_BASE_TM+3).

Version mismatch message: PSI_MT_TM_VERSION_MISMATCH

This type designates a Reactor to PSI Master version mismatch message. It does not contain information besides the message header itself. The version field indicates the next supported version.
It's value is (PSI_MT_BASE_FM+4).

3.7.5.  Node Options

This section describes the constants defined for Node Options. The Node Options are encoded as bit field to conserve bandwidth.

Node Options for messages from the PSI Master:

These Node Options are valid coming from the PSI Master. They may be combined in any number, with certain exceptions. This way it is possible to request all possible information about a Reactor with a single message. Some requests may result in more data to be sent than the current maximum message length. Unless a single Word exceeds the maximum message length (an error; truncation is only possible for non-essential things like vendor-specific information), the reply is split accross multiple messages, using proper sentences.

Query the Reactor's status: PSI_NO_FM_RSREQ

This flag makes the Reactor send a detailed status message to the PSI Master.
It's value is "bit 31".

Query the Reactor's Identification Number: PSI_NO_FM_INREQ

This flag makes the Reactor send a message. Since the IN is part of the message header (see Section 3.5.1 (Message header)), any message that needs to be sent to the PSI Master anyway will do. If no message needs to be sent at that time, an empty message of type PSI_ST_TM_WN is sent.
It's value is "bit 30".

Query vendor specific information: PSI_NO_FM_VSIREQ

This flag is used to request transmission of vendor specific information like vendor name, device name, model number, etc. to the PSI Master. The Reactor sends both the vendor string (using the generic text message Word PSI_ST_TM_WT) and the vendor numbers (using the generic 32 bit Word PSI_ST_TM_WG32), if any. If nothing has been set, an empty sentence of the respective type is sent.
It's value is "bit 29".

Query user specific information: PSI_NO_FM_USIREQ

This flag is used to request transmission of all user specific information. Like PSI_NO_FM_VSIREQ, the Reactor sends both the user string (using the generic text message Word PSI_ST_TM_WT) and the user numbers (using the generic 32 bit Word PSI_ST_TM_WG32), if any. If nothing has been set, an empty sentence of the respective type is sent.
It's value is "bit 28".

Query channel counts: PSI_NO_FM_CCREQ

This flag is used to request transmission of the number of channels of each type (Output, InOut, Input) to the PSI Master. Channels using Safety Sequence count as Output or InOut, according to their properties.
It's value is "bit 27".

Query all data types: PSI_NO_FM_DTREQ

This flag is used to request transmission of the data types of all channels to the PSI Master. If the resulting list is longer than the maximum message length, it may be split into separate messages.
It's value is "bit 26".

Query all channel boundaries: PSI_NO_FM_DBREQ

This flag is used to request transmission of the boundaries of all channels to the PSI Master. If the resulting list is longer than the maximum message length, it may be split into separate messages.
It's value is "bit 25".

Query all channel types: PSI_NO_FM_CTREQ

This flag is used to request transmission of the types of all channels to the PSI Master. If the resulting list is longer than the maximum message length, it may be split into separate messages.
It's value is "bit 24".

Query all channel states: PSI_NO_FM_CSREQ

This flag is used to request transmission of the status of all channels to the PSI Master. If the resulting list is longer than the maximum message length, it may be split into separate messages.
It's value is "bit 23".

Query the Reactor's type: PSI_NO_FM_RTREQ

This flag is used to request transmission of the Reactor's type to the PSI Master.
It's value is "bit 22".

Query all Safe Values: PSI_NO_FM_SVREQ

This flag is used to request transmission of the Safe Values for all channels to the PSI Master. If the resulting list is longer than the maximum message length, it may be split into separate messages.
It's value is "bit 21".

Query all data values: PSI_NO_FM_VREQ

This flag is used to request transmission of the Current Values of all channels to the PSI Master. If the resulting list is longer than the maximum message length, it may be split into separate messages.
It's value is "bit 20".

Query GIDs: PSI_NO_FM_GIDREQ

This flag is used to request transmission of all GIDs assigned to it to the PSI Master.
It's value is "bit 19".

Query maximum message length: PSI_NO_FM_MMLREQ

This flag is used to request transmission of the Reactor's maximum message length to the PSI Master. This SHOULD be at least around 1400 bytes, but MAY be significantly larger. The PSI Master MUST NOT send messages larger than this to the Reactor (including group messages), but may send any size below this. If a Reactor's maximum message size changes, it MUST set the PSI_RS_REREAD_ALL status.
Note that even though the Reactor must know the PSI Master's maximum message length, there is no corresponding request flag. That is because the PSI Master will send this information by itself, usually alongside it's REACTOR_ACCEPTED message. If for some reason the Reactor has not received this information despite a REACTOR_ACCEPTED message, it must resume sending discovery messages.
It's value is "bit 18".

Request to use the Current Values: PSI_NO_FM_GODATA

If this flag is set, and the Reactor is currently using Safe Values, it will start using the Current Values as soon as possible, usually after reception of values for every channel. In any case, a Reactor MUST NOT start using Current Values if a timeout occurred, or if it has not actually received values, or the values received are not valid (anymore). If it is sent in a message containing data Words, then the values received are applied immediately. If the same message has the flag PSI_NO_FM_GOSAFE set, then that MUST be given precedence. PSI_SO_FM_GOSAFE, if set in this message or previously, overrides this flag on a per-channel basis, but does not clear it. Therefore, PSI_NO_FM_GODATA applies to these channels as soon as PSI_SO_FM_GODATA is received, unless it has been cleared meanwhile by PSI_NO_FM_GOSAFE or a timeout occurred. As an exception, channels using Safety Sequence MUST discard received data unless they are set to use Current Values. This flag may be sent at any time, but is restricted to Output and InOut Reactors.
It's value is "bit 17".

Note: because there is no guarantee that any message sent actually reaches the receiver, this flag SHOULD be repeated as long as it is supposed to be in effect. This can easily be done by setting this flag in all Node Sections. Alternatively, the Reactor's status may be queried, or replies that can doubtlessly be attributed to requests having been sent in this same sentence may also be taken as acknowledgement. Furthermore, the flag PSI_NO_FM_SENDACK may also be used.

Request to use the Safe Values: PSI_NO_FM_GOSAFE

If this bit is set, all Output and InOut channels MUST be switched to their respective Safe Values, regardless of the remaining content of the sentence or other Sentence Options. This MUST only be reverted upon reception of PSI_NO_FM_GODATA. This flag MUST be applied regardless of the state of the channels, even if they have already been individually set to use Safe Values. This means that when a channel gets set to use Current Values by use of PSI_SO_FM_GODATA, it will still continue using Safe Values until PSI_NO_FM_GODATA is sent to it's Reactor. Channels using Safety Sequence MUST clear any progress made through their sequence, and cease operation or otherwise ensure safety if at all possible.
This flag SHOULD be used whenever a PSI Master cannot generate new values for any channel of that Reactor, for example if input channels that are used to generate the values cease providing data. It MUST be used if the PSI Master detects an internal or external problem that could result in incorrect values being generated.
It may also be used prior to shutting down the PSI Master, or before a Reactor is removed from the PSI Master's inventory.
It is valid for all Reactor types and may be sent at any time and combined with all other flags; if PSI_NO_FM_GODATA is set then that is ignored, PSI_NO_FM_GOSAFE always taking precedence. Input Reactors ignore this flag.
It's value is "bit 16".

Note: because there is no guarantee that any message sent actually reaches the receiver, this flag must be repeated as long as it is supposed to be in effect. This can easily be done by setting this flag in all Node Sections. Alternatively, the Reactor's status may be queried, or the flag PSI_NO_FM_SENDACK may be used, so a correspondingly set PSI_NO_TM_ACK suffices.

Node Section applies to all Group members: PSI_NO_FM_CLUSTER

This flag is used to indicate that the current Node Section is to be interpreted by all receiving Reactors that are members of the Group specified, and MUST only be used in messages of type broadcast or multicast. A Node Section that has this flag set has no TIN field. This flag exists to increase efficiently in case there are several Reactors that are supposed to receive identical data. Normally, this will be the case for functionally similar Reactors, but because it may be combined with any flag except PSI_NO_FM_ACK, PSI_NO_FM_REACTOR_ACCEPTED and PSI_NO_FM_NODE_FINISHED, it is not limited to sending data (see Section 2.2 (Clusters)). While Reactors MUST implement it, implementation and use of this flag MAY be omitted for PSI Masters.
It's value is "bit 15".

Reply to a Discovery message: PSI_NO_FM_REACTOR_ACCEPTED

If this flag is set, the PSI Master has added the Reactor to it's inventory. The PSI Master sends a message with this flag set for each Discovery Message it receives, until the Reactor ceases sending further Discoveries. This ensures that even if an arbitrary number of Discovery messages and / or replies get lost, every Reactor in the Nexus is known to the PSI Master, and has received a reply, in the end. Therefore, once the PSI Master does not receive any Discovery messages anymore, the initial Discovery Phase has completed (except if a connection is so lossy that no Discoveries reach the PSI Master before the timeout; in this case the Reactor will eventually be discovered as if it had been added later).
It's value is "bit 14".

This means that

• prior to receipt of a reply by a Reactor, the PSI Master has received at least one Discovery message from that Reactor and therefore knows of it and that
• upon arrival of a Discovery message at the PSI Master the sending Reactor may not yet have received a reply, or the Discovery may already have been on it's way before the Reactor received the reply. Since this cannot be known, another reply must be sent. See Section 3.8 (Dynamics) for details.

Request to send a confirmation: PSI_NO_FM_SENDACK

The receiving Reactor MUST acknowledge, after having executed them, all Node Sections that have this flag set. This is done by sending each of them back (keeping the FM types) inside a Node Section that has the flag PSI_NO_TM_ACK set. Only those data and requests must be contained that were actually used and executed.
Doing this may be especially worthwhile in case of channels using Safety Sequence, so the advancement can be monitored.
It's value is "bit 13".

Confirmation of executing a request: PSI_NO_FM_ACK

This flag indicates that the PSI Master has received and executed one or more requests by a Reactor. The Node Section with this flag set contains all requests and data that were executed or used (keeping the TM types). This kind of acknowledge is sent when PSI_NO_TM_SENDACK is set.
It's value is "bit 12".

End of relevant message content: PSI_NO_FM_NODE_FINISHED

If this flag is set in a Node Section it indicates that this is the last Node Section containing data for the specific Reactor. The Reactor therefore does not need to interpret the remainder of the message, if any. However, for error detection, the entire message still needs to be checked, meaning that it must have been received in full and the length data must match each other. If this is not the case, then the message must be discarded and no information or requests contained in it must be used or executed because it is not clear which part the error actually affects.
It's value is "bit 11".

Node Options for messages to the PSI Master:

The following Node Options are valid going to the PSI Master. With certain exceptions, they may be combined arbitrarily.

Reactor status OK: PSI_NO_TM_SOK

If this flag is set, there are no issues at all. It therefore must only be set if none of the other Reactor status bits (PSI_NO_TM_Sx) is set.
It's value is "bit 31".

Non-critical information: PSI_NO_TM_SNCI

If this flag is set, there is non-critical information, so a status inquiry is recommended. The correct functioning of the Reactor is not affected. This is the case, for example, when the Reactor has been switched to Safe Values. If this flag is set, PSI_NO_TM_SOK must not be set.
It's value is "bit 30".

Non-critical error: PSI_NO_TM_SNCE

If this flag is set, there is a non-critical error, so a status inquiry is strongly recommended. This is the case, for example, if the Reactor has started using Safe Values due to a timeout. If this flag is set, PSI_NO_TM_SOK must not be set.
It's value is "bit 29".

Critical error: PSI_NO_TM_SCE

If this flag is set, there is a critical error that may affect the Reactor's operation and possibly damage it. Therefore, a status inquiry is required to query the exact problem(s). This is the case, for example, if channels stop functioning or another hardware or software error has been detected. If this flag is set, PSI_NO_TM_SOK must not be set.
It's value is "bit 28".

Status unchanged: PSI_NO_TM_SUNCH

The Reactor's status is not OK, but has not changed since the last status inquiry; another inquiry is not required, but allowed at any time. If this flag is set, then PSI_NO_TM_SOK must not be set.
It's value is "bit 27".

Request to send a confirmation: PSI_NO_TM_SENDACK

The PSI Master MUST confirm all Node Sections marked with this flag after having executed them. This is done by sending each of them back (keeping the TM types) inside a Node Section that has the flag PSI_NO_FM_ACK set. Only those data and requests must be contained that were actually used and executed.
It's value is "bit 26".

Confirmation of executing a request: PSI_NO_TM_ACK

This flag indicates that the Reactor has received and executed one or more requests by the PSI Master. The Node Section with this flag set contains all requests and data that were executed or used (keeping the FM types). This kind of acknowledge is sent when PSI_NO_FM_SENDACK is set.
It's value is "bit 25".

End of relevant message content: PSI_NO_TM_MASTER_FINISHED

If this flag is set in a Node Section it indicates that this is the last Node Section containing data for the specific PSI Master. The PSI Master therefore does not need to interpret the remainder of the message, if any. However, for error detection, the entire message still needs to be checked, meaning that it must have been received in full and the length data must match each other. If this is not the case, then the message must be discarded and no information or requests contained in it must be used or executed because it is not clear which part the error actually affects.
It's value is "bit 24".

3.7.6.  Sentence Options

This section describes the constants defined for Sentence Options. The Sentence Options are encoded as bit field to conserve bandwidth.
The definitions of channel number sizes apply to all Words within any given sentence. Every node MUST be able cope with all sizes, but the smallest size possible to SHOULD always be used. Especially, if there is a substantial number of channels below any specific size definition in the same sentence as one or more above that size definition, they SHOULD be split into separate sentences so the smaller type can be used for the lower channels.

Sentence Options for messages from the PSI Master:

These Sentence Options are valid coming from the PSI Master. They may be combined in any number, with certain exceptions. This way it is possible to request all possible information about channels using a single message.

Status request: PSI_SO_FM_SREQ

When this bit is set, the Reactor will send the status of the channels given in the sentence. Normally, it is used in sentences of type Channel Number, but it may be used with any type that contains a channel number. It is valid for all channel types and may be sent at any time.
It's value is "bit 31".

Value request: PSI_SO_FM_VREQ

If this bit is set, the Reactor will send the Current Values of the channels given in the sentence (for output channels, the value assigned last is sent). Normally, it is used in sentences of type Channel Number, but it may be used with any type that contains a channel number. It is valid for all channel types and may be sent at any time.
It's value is "bit 30".

Safe Value request: PSI_SO_FM_SVREQ

If this bit is set, the Reactor will send the Safe Values for the channels given in the sentence. Normally, it is used in sentences of type Channel Number, but it may be used with any type that contains a channel number. It is valid only for channels of type Output and InOut, and may be sent at any time. Channels using Safety Sequence send either of the cancellation values given in Section 3.6 (Word Types) at Word type "PSI_Word_Safety_Sequence".
It's value is "bit 29".

Query the data types used: PSI_SO_FM_DTREQ

When this bit is set, the Reactor will send the Data Types (that define the Word type to be used) of the channels given in the sentence. It may be used only in Words of type Channel Number, because the type of Data Word to be used is not known yet. Likewise, the use of the flag PSI_SO_FM_GODATA is not allowed, because for lack of knowledge of the Data Types, no data can have been sent. Use of the flag PSI_SO_FM_GOSAFE is allowed. This Option is valid for all channel types and may be sent at any time. Because a channel using Safety Sequence always uses an additional data Word, that needs to be returned. A channel using Safety Sequence MUST discard any progress through the sequence if this Option is received, and ensure safety if at all possible.
It's value is "bit 28".

Note: Replies to PSI_SO_FM_SVREQ or PSI_SO_FM_VREQ allow deducing the Word Type used, so using this flag is not strictly necessary.

Query Safety Sequence: PSI_SO_FM_SSREQ

When this bit is set, the Reactor will send the Safety Sequences for the channels given in the sentence. Because the sequence is not supposed to be known yet, it may only be used with Words of type Channel Number. It's use is restricted to channels using Safety Sequence and may be sent at any time. A channel using Safety Sequence MUST discard any progress through the sequence if this Option is received, and also cease operation and ensure safety if at all possible.
It's value is "bit 27".

Query channel type: PSI_SO_FM_CTREQ

If this bit is set, the Reactor will send the types of the channels given in the sentence. Because the channel details cannot be known yet, it may only be used with Words of type Channel Number. The use of PSI_SO_FM_GODATA is not allowed in this sentence, since no data values can have been sent yet. Use of PSI_SO_FM_GOSAFE is allowed. It may be used with all channel types and may be sent at any time.
It's value is "bit 26".

Query the data boundaries: PSI_SO_FM_DBREQ

If this bit is set, the Reactor sends the minimum and maximum values that are meaningful to the channels listed in the sentence. This bit may be used with any Word type that contains a channel number, and may be sent to channels of all types and at any time. Even though use of this flag is technically optional, and it is expected that for most channels the boundaries will be given by the Word type, it should be used because that is not necessarily the case. Therefore it is important for correctly calculating values to send, especially since a receiver may use any method for fitting the data received into it's storage (discarding the superfluous bits is suggested).
It's value is "bit 25".

Defines channel numbers to be 8 bit: PSI_SO_FM_CN8

This bit defines the size of all channel numbers in the sentence to be 8 bit in size. It may only be used in conjunction with Words actually containing channel numbers, and must not be combined with any other channel number definition.
It's value is "bit 24".

Defines channel numbers to be 16 bit: PSI_SO_FM_CN16

This bit defines the size of all channel numbers in the sentence to be 16 bit in size. It may only be used in conjunction with Words actually containing channel numbers, and must not be combined with any other channel number definition.
It's value is "bit 23".

Defines channel numbers to be 32 bit: PSI_SO_FM_CN32

This bit defines the size of all channel numbers in the sentence to be 32 bit in size. It may only be used in conjunction with Words actually containing channel numbers, and must not be combined with any other channel number definition.
It's value is "bit 22".

Defines channel numbers to be 64 bit: PSI_SO_FM_CN64

This bit defines the size of all channel numbers in the sentence to be 64 bit in size. It may only be used in conjunction with Words actually containing channel numbers, and must not be combined with any other channel number definition.
It's value is "bit 21".

Maximum message length information: PSI_SO_FM_MMSINFO

This flag is valid only in conjunction with a Word of type PSI_Word_Node_Specification, that contains the maximum message length of the PSI Master. A Reactor MUST NOT send messages larger than that to the PSI Master, but may send any size below. The PSI Master MAY change this at any time during normal operation, usually as part of adapting to packet loss. It then sends a message with this flag, and usually also the flag PSI_NO_FM_SENDACK set. This way, and because the PSI Master controls the length of it's own messages, it is in full control of the maximum message sizes that are sent, and therefore the Reactors do not need to be reconfigured to adapt to new network conditions.
The PSI Master's maximum message length SHOULD be around 1400 bytes, at least initially, to allow configuration information to be sent as complete as possible, but MAY become significantly larger and also smaller in response to network packet loss.
This means that Words without fixed length, like vendor- and user-specific information, should be well below 1400 bytes in length, because otherwise they may be truncated, or sending may abort with an error. Don't be unnecessarily chatty! ;)
It's value is "bit 20".

Assignment of Group IDs: PSI_SO_FM_GIDSET

This bit may only be used in conjunction with Words of type PSI_Word_Node_Specification, that contain the GIDs to be used. Multiple Words may be used to assign as many GIDs. Any sentence having this flag set clears and replaces all GIDs assigned previously, if any. Use of this flag in an empty sentence removes all assigned GIDs. Execution must be acknowledged (see PSI_NO_TM_ACK). Use of this flag is optional.
It's value is "bit 19".

Additional assignment of Group IDs: PSI_SO_FM_GIDADD

This bit may only be used in conjunction with Words of type PSI_Word_Node_Specification, that contain the GIDs to be added. Multiple Words may be used to add as many GIDs. Any sentence having this flag set adds the GIDs listed to any GIDs that were assigned previously; how to handle duplicates is left to the implementor. Execution must be acknowledged (see PSI_NO_TM_ACK). Use of this flag is optional.
It's value is "bit 18".

Removal of Group IDs: PSI_SO_FM_GIDREM

This bit may only be used in conjunction with Words of type PSI_Word_Node_Specification, that contain the GIDs to be removed. Multiple Words may be used to remove as many GIDs. Any duplicates must also be removed from the list. Use of this flag in an empty sentence removes all assigned GIDs. Execution must be acknowledged (see PSI_NO_TM_ACK). Use of this flag is optional.
It's value is "bit 17".

Request to use Current Values: PSI_SO_FM_GODATA

If this flag is set, and the channels listed are currently using Safe Values, they will start using the Current Values as soon as possible, usually after reception of the first value. In any case, a channel MUST NOT start using Current Values if a timeout occurred, or if it has not actually received values, or the values received are not valid (anymore). This flag may be used with any Word type containing a channel number, and at any time, but is restricted to Output and InOut channels. If it is set in a sentence containing data Words, then the values received are applied immediately. If the same sentence has the flag PSI_SO_FM_GOSAFE set, then that MUST be given precedence. PSI_NO_FM_GOSAFE, if set in this message or previously, overrides this flag, but does not clear it. Therefore, PSI_SO_FM_GODATA is applied as soon as PSI_NO_FM_GODATA is received, unless it has been cleared meanwhile by PSI_SO_FM_GOSAFE or a timeout occurred. As an exception, channels using Safety Sequence MUST discard received data unless they are set to use Current Values.
It's value is "bit 16".

Note: because there is no guarantee that any message sent actually reaches the receiver, this flag SHOULD be repeated as long as it is supposed to be in effect. This can easily be done by setting this flag in all sentences. Alternatively, the channels status may be queried, or replies that can doubtlessly be attributed to requests having been sent in this same sentence may also be taken as acknowledgement. Furthermore, the flag PSI_NO_FM_SENDACK may also be used.

Request to use Safe Values: PSI_SO_FM_GOSAFE

If this bit is set, all channels listed in this sentence MUST be switched to their respective Safe Values, regardless of the remaining content of the sentence or other Sentence Options. This MUST only be reverted upon reception of PSI_SO_FM_GODATA. This flag MUST be applied regardless of the state of the Reactor, even if the Reactor as a whole has already been set to use Safe Values. This means that when the Reactor gets set to use Current Values by use of PSI_NO_FM_GODATA, channels having received this flag continue using Safe Values until PSI_SO_FM_GODATA is sent to them. Channels using Safety Sequence MUST clear any progress made through their Safety Sequence, and cease operation or otherwise ensure safety if at all possible. This flag MUST be used whenever a PSI Master cannot generate new values for a channel, for example if an Input channel that is used to generate the value ceases providing data. It is valid for all channel types (Input channels ignore it) and may be sent at any time and combined with all other flags; if PSI_SO_FM_GODATA is set then that is ignored, PSI_SO_FM_GOSAFE always taking precedence.
It's value is "bit 15".

Note: because there is no guarantee that any message sent actually reaches the receiver, this flag must be repeated as long as it is supposed to be in effect. This can easily be done by setting this flag in all sentences. Alternatively, the channel's status may be queried, or the flag PSI_NO_FM_SENDACK may be used, so a correspondingly set PSI_NO_TM_ACK suffices.

Assignment of new values: PSI_SO_FM_VSET

This flag must only be used in sentences containing data Words and makes the Reactor assign the new values to the corresponding channels. Except for channels using Safety Sequence, this applies regardless of the current state of the Reactor: the values are applied even if the Reactor or channels are set to use Safe Values. As soon as these conditions are removed, the values are used as long as no timeout occurs. Channels using Safety Sequence MUST discard any values sent to them unless they are set to use Current Values and have no other prohibitive states. This flag is valid for Output and InOut channels only.
It's value is "bit 14".

Sentence Options for messages to the PSI Master:

These Sentence Options are valid in messages sent to the PSI Master. Contrary to those coming from the PSI Master, and except for being combined with a channel number definition, they are mutually exclusive because they define the meaning of generic Words.

Words contain Current Values: PSI_SO_TM_VINFO

This flag means that the Words in this sentence contain the Current Values of the respective channels.
It's value is "bit 31".

Words contain Safe Values: PSI_SO_TM_SVINFO

This flag means that the Words in this sentence contain the Safe Values for the respective channels. A Reactor must only send these values when requested, to minimize chances for erroneously applying incorrect values. Therefore the PSI Master must check that there actually are pending requests (PSI_SO_FM_SVREQ) for the channels listed, that have not yet timed out, and otherwise discard the values.
It's value is "bit 30".

Words contain Data Type information: PSI_SO_TM_DTINFO

This flag means that the Words in this sentence contain the data types used by the channels listed. These need to be known before values can be sent or received.
It's value is "bit 29".

Words contain Channel Type information: PSI_SO_TM_CTINFO

This flag means that the Words in this sentence contain the types of the channels listed. These need to be known before values can be sent or received.
It's value is "bit 28".

Words contain data boundary information: PSI_SO_TM_DBINFO

This flag means that the Words in this sentence contain the boundaries of the channels listed. These should be known before values are sent or received.
In order to prevent proliferation of Word types, the same types as used for data values are used. Therefore for each channel, two data Words of the same type are sent consecutively within the same sentence, both containing the channel number. The first Word contains the minimum value and the second one contains the maximum value in the value syllable. The overhead created by sending the channel number a second time is tolerable because this kind of sentence is used only upon initialization of the Nexus and thus does not impact normal operation.
It's value is "bit 27".

Defines channel numbers to be 8 bit: PSI_SO_FM_CN8

This bit defines the size of all channel numbers in the sentence to be 8 bit in size. It may only be used in conjunction with Words actually containing channel numbers, and must not be combined with any other channel number definition.
It's value is "bit 26".

Defines channel numbers to be 16 bit: PSI_SO_TM_CN16

This bit defines the size of all channel numbers in the sentence to be 16 bit in size. It may only be used in conjunction with Words actually containing channel numbers, and must not be combined with any other channel number definition.
It's value is "bit 25".

Defines channel numbers to be 32 bit: PSI_SO_TM_CN32

This bit defines the size of all channel numbers in the sentence to be 32 bit in size. It may only be used in conjunction with Words actually containing channel numbers, and must not be combined with any other channel number definition.
It's value is "bit 24".

Defines channel numbers to be 64 bit: PSI_SO_TM_CN64

This bit defines the size of all channel numbers in the sentence to be 64 bit in size. It may only be used in conjunction with Words actually containing channel numbers, and must not be combined with any other channel number definition.
It's value is "bit 23".

Maximum message length information: PSI_SO_TM_MMSINFO

This flag is valid only in conjunction with a Word of type PSI_Word_Node_Specification, that contains the maximum message length of the Reactor. A PSI Master MUST NOT send messages larger than that to the Reactor, but may send any size below. Even though not normally necessary, the Reactor MAY change this during normal operation, usually after user configuration changes. In that case, it MUST set the status PSI_RS_REREAD_ALL so the PSI Master will request the information anew.
The maximum message length SHOULD be at least around 1400 bytes, but MAY be significantly larger.
It's value is "bit 22".

Word contains Reactor type: PSI_SO_TM_RTINFO

This flag means that the Word in this sentence contains the type of the Reactor. This needs to be known before values can be sent or received.
It's value is "bit 21".

Words contain Group information: PSI_SO_TM_GIDINFO

This flag means that the Words in this sentence contain the GIDs that currently are assigned to the Reactor. These need to be known before broadcast or multicast messages can be sent.
It's value is "bit 20".

Words containing vendor specific information: PSI_SO_TM_VSINFO

This flag means that the Words in this sentence contain vendor specific information; the sentence type is either PSI_ST_TM_WT or PSI_ST_TM_WT.
It's value is "bit 19".

Words containing user specific information: PSI_SO_TM_USINFO

This flag means that the Words in this sentence contain user specific information; the sentence type is either PSI_ST_TM_WT or PSI_ST_TM_WT.
It's value is "bit 18".

3.7.7.  Sentence types

The sentence type defines the type of the Words that the sentence consists of.

Sentence types for messages from a PSI Master:

Sentence of boolean Data Words: PSI_ST_FM_WDB

This type identifies a sentence consisting of Words of type PSI_Word_DataB.
It's value is (PSI_ST_BASE_FM+0).

Sentence of signed 8 bit Data Words: PSI_ST_FM_WDS8

This type identifies a sentence consisting of Words of type PSI_Word_DataS8.
It's value is (PSI_ST_BASE_FM+1).

Sentence of unsigned 8 bit Data Words: PSI_ST_FM_WDU8

This type identifies a sentence consisting of Words of type PSI_Word_DataU8.
It's value is (PSI_ST_BASE_FM+2).

Sentence of signed 16 bit Data Words: PSI_ST_FM_WDS16

This type identifies a sentence consisting of Words of type PSI_Word_DataS16.
It's value is (PSI_ST_BASE_FM+3).

Sentence of unsigned 16 bit Data Words: PSI_ST_FM_WDU16

This type identifies a sentence consisting of Words of type PSI_Word_DataU16.
It's value is (PSI_ST_BASE_FM+4).

Sentence of signed 32 bit Data Words: PSI_ST_FM_WDS32

This type identifies a sentence consisting of Words of type PSI_Word_DataS32.
It's value is (PSI_ST_BASE_FM+5).

Sentence of unsigned 32 bit Data Words: PSI_ST_FM_WDU32

This type identifies a sentence consisting of Words of type PSI_Word_DataU32.
It's value is (PSI_ST_BASE_FM+6).

Sentence of signed 64 bit Data Words: PSI_ST_FM_WDS64

This type identifies a sentence consisting of Words of type PSI_Word_DataS64.
It's value is (PSI_ST_BASE_FM+7).

Sentence of unsigned 64 bit Data Words: PSI_ST_FM_WDU64

This type identifies a sentence consisting of Words of type PSI_Word_DataU64.
It's value is (PSI_ST_BASE_FM+8).

Sentence of floating-point Data Words: PSI_ST_FM_WDFP

This type identifies a sentence consisting of Words of type PSI_Word_DataFP.
It's value is (PSI_ST_BASE_FM+9).

Sentence of double floating-point Data Words: PSI_ST_FM_WDDFP

This type identifies a sentence consisting of Words of type PSI_Word_DataDFP.
It's value is (PSI_ST_BASE_FM+10).

Sentence of quadruple floating-point Data Words: PSI_ST_FM_WDQFP

This type identifies a sentence consisting of Words of type PSI_Word_DataQFP.
It's value is (PSI_ST_BASE_FM+11).

Sentence of one opaque Data Word with variable length: PSI_ST_FM_WDO

This type identifies a sentence consisting of at most one Word with variable length of type PSI_Word_Data_Opaque.
It's value is (PSI_ST_BASE_FM+12).

Sentence of one plain Data Word with variable length: PSI_ST_FM_WDP

This type identifies a sentence consisting of at most one Word with variable length of type PSI_Word_Data_Plain.
It's value is (PSI_ST_BASE_FM+13).

Sentence of generic 32 bit Data Words: PSI_ST_FM_WG32

This type identifies a sentence consisting of Words of type PSI_Word_Generic_32.
It's value is (PSI_ST_BASE_FM+14).

Sentence without content: PSI_ST_FM_WN

This type identifies a sentence containing no Words. It's main purpose is for debugging.
It's value is (PSI_ST_BASE_FM+15).

Sentence of Channel Numbers: PSI_ST_FM_WCN

This type identifies a sentence consisting of Words of type PSI_Word_Channel_Number. It' value is (PSI_ST_BASE_FM+16).

Sentence of one Text Data Word with variable length: PSI_ST_FM_WT

This type identifies a sentence consisting of at most one Word with variable length of type PSI_Word_Text.
It's value is (PSI_ST_BASE_TM+17).

Sentence types for messages to a PSI Master:

Sentence of boolean Data Words: PSI_ST_TM_WDB

This type identifies a sentence consisting of Words of type PSI_Word_DataB.
It's value is (PSI_ST_BASE_TM+0).

Sentence of signed 8 bit Data Words: PSI_ST_TM_WDS8

This type identifies a sentence consisting of Words of type PSI_Word_DataS8.
It's value is (PSI_ST_BASE_TM+1).

Sentence of unsigned 8 bit Data Words: PSI_ST_TM_WDU8

This type identifies a sentence consisting of Words of type PSI_Word_DataU8.
It's value is (PSI_ST_BASE_TM+2).

Sentence of signed 16 bit Data Words: PSI_ST_TM_WDS16

This type identifies a sentence consisting of Words of type PSI_Word_DataS16.
It's value is (PSI_ST_BASE_TM+3).

Sentence of unsigned 16 bit Data Words: PSI_ST_TM_WDU16

This type identifies a sentence consisting of Words of type PSI_Word_DataU16.
It's value is (PSI_ST_BASE_TM+4).

Sentence of signed 32 bit Data Words: PSI_ST_TM_WDS32

This type identifies a sentence consisting of Words of type PSI_Word_DataS32.
It's value is (PSI_ST_BASE_TM+5).

Sentence of unsigned 32 bit Data Words: PSI_ST_TM_WDU32

This type identifies a sentence consisting of Words of type PSI_Word_DataU32.
It's value is (PSI_ST_BASE_TM+6).

Sentence of signed 64 bit Data Words: PSI_ST_TM_WDS64

This type identifies a sentence consisting of Words of type PSI_Word_DataS64.
It's value is (PSI_ST_BASE_TM+7).

Sentence of unsigned 64 bit Data Words: PSI_ST_TM_WDU64

This type identifies a sentence consisting of Words of type PSI_Word_DataU64.
It's value is (PSI_ST_BASE_TM+8).

Sentence of floating-point Data Words: PSI_ST_TM_WDFP

This type identifies a sentence consisting of Words of type PSI_Word_DataFP.
It's value is (PSI_ST_BASE_TM+9).

Sentence of double floating-point Data Words: PSI_ST_TM_WDDFP

This type identifies a sentence consisting of Words of type PSI_Word_DataDFP.
It's value is (PSI_ST_BASE_TM+10).

Sentence of quadruple floating-point Data Words: PSI_ST_TM_WDQFP

This type identifies a sentence consisting of Words of type PSI_Word_DataQFP.
It's value is (PSI_ST_BASE_TM+11).

Sentence of one opaque Data Word with variable length: PSI_ST_TM_WDO

This type identifies a sentence consisting of at most one Word with variable length of type PSI_Word_Data_Opaque.
It's value is (PSI_ST_BASE_TM+12).

Sentence of one plain Data Word with variable length: PSI_ST_TM_WDP

This type identifies a sentence consisting of at most one Word with variable length of type PSI_Word_Data_Plain.
It's value is (PSI_ST_BASE_TM+13).

Sentence of generic 32 bit Data Words: PSI_ST_TM_WG32

This type identifies a sentence consisting of Words of type PSI_Word_Generic_32.
It's value is (PSI_ST_BASE_TM+14).

Sentence without content: PSI_ST_TM_WN

This type identifies a sentence containing no Words. It's main purpose is for debugging.
It's value is (PSI_ST_BASE_TM+15).

Sentence of Channel Numbers: PSI_ST_TM_WCN

This type identifies a sentence consisting of Words of type PSI_Word_Channel_Number.
It's value is (PSI_ST_BASE_TM+16).

Sentence of one Text Data Word with variable length: PSI_ST_TM_WT

This type identifies a sentence consisting of at most one Word with variable length of type PSI_Word_Text.
It's value is (PSI_ST_BASE_TM+17).

Sentence of one Reactor status Word of variable length: PSI_ST_TM_WRS

This type identifies a sentence consisting of at most one Word with variable length of type PSI_Word_Reactor_Status.
It's value is (PSI_ST_BASE_TM+18).

Sentence of channel status Words: PSI_ST_TM_WCS

This type identifies a sentence consisting of Words of type PSI_Word_Channel_Status.
It's value is (PSI_ST_BASE_TM+19).

Sentence of one Safety Sequence Word of variable length: PSI_ST_TM_WSS

This type identifies a sentence consisting of at most one Word with variable length of type PSI_Word_Safety_Sequence.
It's value is (PSI_ST_BASE_TM+20).

Sentence of Channel Counts: PSI_ST_TM_WCC

This type identifies a sentence consisting of Words of type PSI_Word_Channel_Count.
It's value is (PSI_ST_BASE_TM+21).

Sentence of generic Node Specification Words: PSI_ST_TM_W_N_S

This type identifies a sentence consisting of Words of type PSI_Word_Node_Specification.
It's value is (PSI_ST_BASE_TM+22).

Sentence of generic Channel Specification Words: PSI_ST_TM_W_C_S

This type identifies a sentence consisting of Words of type PSI_Word_Channel_Specification.
It's value is (PSI_ST_BASE_TM+23).

3.7.8.  Detailed Reactor Status

This section describes the detailed status as sent via PSI_Word_Reactor_Status by the Reactor.

Reactor status OK: PSI_RS_OK

This constant means that there are no issues. It must not be combined with others.
It's value is 0.

Reactor uses Safe Values due to lack of data: PSI_RS_SAFE

This constant means that the PSI Master has not yet sent a value for all channels, so the Reactor still uses Safe Values for these channels. All other channels are already using the Current Values.
It's value is 1.

Reactor uses Safe Values due to a timeout: PSI_RS_TO_SAFE

This constant means that no values have been received for too long, resulting in a timeout: the Reactor has switched to Safe Values. The PSI Master must send PSI_NO_FM_GODATA to the Reactor to change this status.
It's value is 2.

Reactor uses Safe Values due to an overshoot: PSI_RS_OS_SAFE

This constant means that values have been received that are outside a channel's boundaries: the Reactor has switched to Safe Values. The PSI Master must send PSI_NO_FM_GODATA to the Reactor to change this status.
It's value is 3.

Reactor uses Safe Values due to request: PSI_RS_SET_SAFE

This constant means that the PSI Master has requested the Reactor to switch it's channels to Save Values (see PSI_NO_FM_GODATA). If no new values arrive, a timeout will occur in addition. The PSI Master must send PSI_NO_FM_GODATA to change this status.
It's value is 4.

Software error occurred: PSI_RS_SWFAIL

This constant means that the Reactor has encountered a software problem that may affect it's functionality. This does not include problems in channels.
It's value is 5.

Hardware error occurred: PSI_RS_HWFAIL

This constant means that the Reactor has encountered a hardware problem that may affect it's functionality. This does not include problems in channels.
It's value is 6.

Problem with some channels: PSI_RS_CHANNELS

This constant means that one or more channel is faulty. The PSI Master needs to query each channel's status to see which channels report which problems.
It's value is 7.

Overheat condition: PSI_RS_OVERHEAT

This constant means that the Reactor has overheated and reduced function as much as possible. If the condition is allowed to persist, the Reactor MAY optionally develop hardware damage or switch off, so communication may cease at any time.
It's value is 8.

PSI Master uses incorrect data types: PSI_RS_REREAD_DT

At least one channel is being fed using an incorrect data type so the data cannot be used. The PSI Master must correct the problem, possibly requiring (re-)reading of the data types.
It's value is 9.

Reactor configuration changed: PSI_RS_REREAD_ALL

This constant means that changes have been made to the Reactor that may have invalidated information previously queried. Therefore the Reactor has switched to Safe Values, and the PSI Master must re-read all information. This may occur, for example, when channels are added, removed or swapped. As long as the PSI Master has not queried the data anew, no data are accepted.
It's value is 10.

3.7.9.  Channel status

This section describes channel status values. They are binary constants intended to be put into the channel status bit field of the Word PSI_Word_Channel_Status. If the status of a channel changes to anything non-OK, then the Reactor status must reflect this by adding PSI_RS_CHANNELS. Also, if there is a change, the flag PSI_NO_TM_SUNCH must be cleared.

Channel status OK: PSI_CS_OK

If this flag is set, then the channel works normally.
It's value is "bit 15".

The channel uses Safe Value due to lack of data: PSI_CS_SAFE

If this flag is set, then the channel has not yet been fed a value and therefore is using it's Save Value.
It's value is "bit 14".

The channel uses Safe Value due to timeout: PSI_CS_TO_SAFE

If this flag is set, then the channel has not been fed data for a while, so a timeout occurred and the channel has switched to it's Safe Value.
It's value is "bit 13".

The channel uses Safe Value due to value overshoot: PSI_CS_OS_SAFE

If this flag is set, then the channel has been fed values outside it's boundaries, so it has switched to it's Safe Value. It's value is "bit 12".

The channel uses Safe Value due to request: PSI_CS_SET_SAFE

If this flag is set, then the channel has been switched to it's Safe Value. It's value is "bit 11".

Software error: PSI_CS_SWFAIL

If this flag is set, then the channel is faulty due to a software problem.
It's value is "bit 10".

Hardware error: PSI_CS_HWFAIL

If this flag is set, then the channel is faulty due to a hardware problem.
It's value is "bit 9".

Overheat condition: PSI_CS_OVERHEAT

If this flag is set, then the channel has overheated and reduced functionality as much as possible. If the condition persists, the channel may develop hardware failure.
It's value is "bit 8".

Channel does not exist: PSI_CS_VIRTUAL

If this flag is set, then the channel does not physically exist. This is no error: it merely indicates that no physical channel has been assigned to a logical one.
It's value is "bit 7".

Channel status unknown: PSI_CS_UNKNOWN

If this flag is set, then the channel hardware has no way of checking it's functioning, so it's status is not known. If this is the case, PSI_CS_OK MUST NOT be returned.
It's value is "bit 6".

3.7.10.  Reactor types

This section describes the constants for the different Reactor types.

Output Reactor: PSI_RT_OUTPUT

This constant means that the Reactor possesses only Output channels, not generating any data. This is the most common Reactor type like dimmers single spots LASER systems, pyro effects, etc.. Output channels always allow reading of their Current Values.
It's value is 0.

Input Reactor: PSI_RT_INPUT

This constant means that the Reactor possesses only Input channels, generating data but not consuming any. Reactors of this kind include touch boards, fader panels, etc..
It's value is 1.

Combined Input and Output Reactor: PSI_RT_INOUT

This constant means that the Reactor possesses both Input and Output channels. Reactors of this type may be non-transparent bidirectional protocol converters, sensor/actor combinations, etc.. Output Reactors always allow reading back their values, so they do not belong to this type. Reactors having only channels of type PSI_CT_NONE also do not belong to this type.
It's value is 2.

Reactor without function: PSI_RT_NONE

This constant means that the Reactor possesses only channels of type PSI_CT_NONE or no channels at all. Reactors of this type may be testing equipment like protocol analyzers that are able to connect to a PSI Master but do not generate or consume channel data.
It's value is 3.

3.7.11.  Channel types

This section describes the constants representing the channel types.

Output channel: PSI_CT_OUTPUT

This constant means that the channel consumes data and outputs them in some way. Most channels are of this type, like dimmer channels or control channels of foggers. Channels of this type always allow reading out their Current Values.
It's value is 0.

Input channel: PSI_CT_INPUT

This constant means that the channel creates data, for example in a fader panel.
It's value is 1.

Combined Input- and Output channel: PSI_CT_INOUT

This constant means that the channel both creates and consumes data. Because it also outputs values, it MUST have a Safe Value that can be read by the PSI Master in the usual way. Because pure Output channels always allow reading out their Current Values, they do not belong to this type.
It's value is 2.

Channel without function: PSI_CT_NONE

This constant means that the channel has no function, and can neither consume nor create data. This type may be used for testing; logical channels that are not assigned to a physical channels may also have this type. Data sent to channels of this kind will be discarded, queries for Safe Values, etc. get ignored.
It's value is 3.

Channel using Safety Sequence: PSI_CT_SAFETY

This constant means that the channel controls something that requires extra safety measures; erroneous changes must be avoided at nearly all costs. This may be a pyro effect, a LASER controller, high-voltage effects, motors, etc.. It requires correct reception of a sequence of 32 bit values to accept data. The sequence required may be queried using PSI_SO_FM_SSREQ. Any Reactor containing channels of this type MUST ensure safety if at all possible whenever it receives a query about it's properties (like data type and even vendor information), because that indicates that the PSI Master does not (anymore) know how to access that channel, and therefore whatever value it currently uses may be invalid and thus potentially dangerous.
It's value is 4.

3.8.  Dynamics

This section describes the dynamical behavior of PSI. Only those components relevant to the discussion are looked at; for a complete list, refer to Section 3 (Design).

Note that if a message requests information, the reply must be sent as soon as possible: the node should not wait for possible further requests that might allow conglomeration of replies. However, if there already are replies that have not yet been sent, then the node may do such conglomeration. In any case, the replies generated last MUST be placed behind the previous ones, so that the sequence will always be retained. This is especially important if for some reason values for the same channel are put into the same message, since otherwise outdated values would be mistaken for the most recent ones.
If the same request is received more than once, and the previous requests have not yet been processed, then only the most recent one needs to be processed. However, a node must not wait to see if there may be duplicate requests.

The periodic status inquiries are an exception to the above: the PSI Master may delay sending of these requests for some tenths of seconds if no other messages, to which the requests could be added, are ready to be sent. However, the Reactors must still reply immediately.

3.8.1.  Discovery Phase

The use of Reactor-specific channel numbers and the resulting explicit addressing of individual Reactors necessitates that the PSI Master be informed about every single Reactor in the Nexus. Since, however, the Reactors also do not know about the PSI Master, neither side can contact the other directly to create that knowledge.
Because of that, the broadcast mechanism is used. After the PSI Master starts up, it will send messages of type PSI_MT_FM_DISCOVERY for a while (this SHOULD be between two seconds and one minute, but implementations MAY allow the user to change these settings). During this time, it does not reply to any messages coming from the Reactors. The large number of messages sent ensures that every Reactor that can be reached will receive at least one. Through reception of the Discovery message from the PSI Master, Reactors are informed that the PSI Master does not yet know about them, so they must inform it about themselves. Because the Discovery message contains the PSI Master's IN, the Reactors can identify the specific PSI Master in case there are more than one; the PSI Master's IP address is delivered as well, so it can be used for subsequent messages.

Upon reception of a message of type PSI_MT_FM_DISCOVERY, a Reactor MUST start sending messages of type PSI_MT_TM_DISCOVERY. These messages are sent to the PSI Master as normal unicast, until it acknowledges their reception. To avoid having one Reactor clog the network, a Reactor SHOULD wait between 10 and 100 milliseconds between sending each such message.

After sending the Discovery messages, the PSI Master SHOULD wait between two and ten seconds (it MAY allow the user to configure other times), while it receives and acknowledges replies from Reactors. Because a possibly huge amount of messages may have been generated, resulting in massive packet loss, this timeout is reset whenever a reply is received. This way, all Reactors will eventually be acknowledged. The Discovery Phase ends when no replies have been received for the specified time.
The acknowledgements sent by the PSI Master are messages of type PSI_MT_FM_NORMAL, PSI_MT_FM_MULTICAST or PSI_MT_FM_BROADCAST, containing at least one Node Section that has the flag PSI_NO_FM_REACTOR_ACCEPTED set (this MUST be set in the first Node Section for the respective Reactor). By using the multicast or broadcast types, multiple Reactors can be acknowledged using a single message. Additionally, all elements that are available during normal operation may be used as well, so the acknowledgement message may be used to request information or to send data. It must be noted, however, that due to the remaining Reactors that have not yet been acknowledged and therefore are still sending Discovery messages, messages may get dropped. Therefore, it usually is advisable to only request information after the Discovery Phase has finished.
Network latency can make the PSI Master receive a Reactor's Discovery message after the Reactor has received an acknowledgement by the PSI Master. Because the PSI Master can not know if the Reactor actually received the acknowledge or not, it must send an acknowledge whenever it receives a Discovery message. The Reactor must therefore be prepared to handle this, for example by discarding previous or excess acknowledgements.

The result is "at least once" semantics. Every message is idempotent, so that the entire process can be restarted by either side at any time, while the number of (identical) messages received by either side has no influence on the outcome.

Because the PSI Master ceases sending Discovery messages before starting to acknowledge any Reactors, and alternate routes should not exist, a Reactor will not normally receive a Discovery message after an acknowledgement. Regardless, every Reactor MUST start sending Discovery messages whenever it receives a Discovery message from a PSI Master. This is required because the PSI Master may have been restarted, or the Discovery Phase may have been triggered for other reasons.

Whenever a Reactor is (re)started, it MUST begin sending Discovery messages in a similar way as when it has received a Discovery message from a PSI Master. Since the Reactor does not yet know the PSI Master, it must send these messages as broadcast. Whenever a PSI Master receives such a message, it MUST acknowledge it in the normal way. By doing so, Reactors that appear during normal operation can be taken care of.
Whenever a Reactor has not received any messages from the PSI Master for at most two minutes (this may be user-configurable), it may try sending unicast Discovery messages to the IP address last used by the PSI Master. If that is not known, or the PSI Master does not reply during that time, the Reactor MUST revert to sending broadcast Discovery messages, because the PSI Master's IP address may have changed (the unicast step MAY be skipped). This may happen when the PSI Master has been restarted, and completed it's Discovery Phase, while it or the Reactor were disconnected from the network. This method allows for that, so the Reactor will be taken care of after the connection has been reestablished.

This way, all nodes can be relatively certain that, in case one gets restarted, the respective partner knows that the connection needs to be recreated. It cannot be 100% reliably be ensured if one takes into account that restarts may get masked by certain combinations of cabling disconnects and duplicated, old packets, but even then there is no real problem, because the IN is unique. If two or more Reactors receive swapped IP addresses after a restart, and the PSI Master erroneously sends messages to the wrong Reactor, this will be noticed by the receiving Reactor because it's IN does not match the TIN from the message. In the simplest case, it will just ignore the entire message, so the PSI Master will not receive any replies and regard the Reactor as lost. The erroneously contacted Reactor MAY also start sending Discovery messages to the PSI Master after a while to alert it of the swap, or the PSI Master MAY send Discovery messages to Reactors that don't respond. In any case it is ensured that no incorrect data are used.

Reactors do not send a reply after receiving the acknowledgement from the PSI Master; they just cease sending Discovery messages. This means that the PSI Master does not know if the Reactor has received the acknowledgement, or if it was disconnected or has failed. However, this information is automatically gathered later, by way of the periodic status inquiries.



Note: whenever a Reactor sends Discovery messages, it MUST conform to the interval mentioned above.

3.8.2.  Initialization Phase

After the Discovery Phase has completed, an Initialization Phase starts. It is used to query information about all Reactors that have been discovered.
Every Reactor is characterized by a number of properties describing it's function and uses. The PSI Master sends a number of queries to the Reactor to collect this information. All such information may be queried in a single message, or be distributed across several messages, possibly containing data sent to the Reactor. The sequence of the queries may be chosen arbitrarily, as well as whether to wait for each reply before sending the next query or not. Also, as described in Section 3.7 (Constants), some need not be queried in all cases. The individual queries are:

1. Reactor type
2. Reactor status
3. vendor specific information
4. user specific information
5. channel counts
6. channel types
7. data formats
8. data boundaries
9. maximum message length
10. Safe Values
11. channel status

Especially with Reactors having many channels, replies to one or more queries may need to be split into multiple messages. It is within the Reactor's discretion to decide in which way to do this, so parts of different replies may be conglomerated into a single message. The PSI Master must thus be prepared to re-request any part of the required information that may have been lost in transit.

Reactors MUST reply to all of these requests, but may decide wether to combine multiple replies or not.

When all information is known to the PSI Master, the Initialization Phase has completed.

Note that Reactors that appear during normal operation MUST also be queried in this way. The PSI Master MAY flag them in any way convenient.

3.8.3.  Normal Operation

Periodic status inquiries

In order to always be informed about the status of the Reactors in the Nexus, the PSI Master must query their status periodically. The interval SHOULD NOT be made smaller than about 1 to 2 Hz, because the information is evaluated by the user, resulting in much larger latency, but SHOULD be configurable. This way, the network is not overly burdened by these inquiries. Since these queries can easily be added to messages that need to be sent regardless, the overhead can be further reduced. Since nearly all Reactors are expected to receive data or queries at a much smaller interval, almost no extra messages will need to be sent to accommodate this. On the other hand, a Reactor MUST NOT delay replies to status inquiries in order to combine them with another message, but MAY do so, within length limits, if there is a message pending for sending already. This may be convenient if pre-arbitration (not discussed in this memo) is employed to schedule messages sent to the PSI Master. If no message is pending for sending, a dedicated message must be sent.

Data messages

This is the most common form of communication, as data messages are the core intention of this protocol. The goal is to reach the highest refresh rate possible, so data must be sent as soon as possible. Especially, a node must not wait for data to be generated before sending what has already been generated. This does however not mean that a message must be sent for each channel of a Reactor. Actually, a node must strive to minimize the number of messages sent by packing data for as many channels as possible into every message, for example by arranging data generation or consumption in a way that ensures data can be sent and received at any time. However, the maximum message length may be dynamically adjusted to reach the lowest net data loss if the network proves unreliable. In that case, the node may distribute data for any Reactor's channels across multiple messages, but keeping identical data logically ordered is still sensible. This is because using a single Word type message saves the overhead of requiring multiple Sentence Headers, reducing message size and therefore probability for errors, as well as bandwidth needed. In such a case, multicast and broadcast messages also should contain data for fewer nodes than theoretically possible, in favor of smaller size. Loss feedback methods are not discussed as part of this memo, however. An implementation MAY allow the user to set a maximum message length that best suits their network.
A data message itself does not require a reply, but may be combined with requests.

Data requests

If no pre-arbitration is performed, data generated by Input and InOut Reactors must be explicitly requested by the PSI Master (polling). As before, the objective is a refresh rate that is as high as possible. Therefore it is imperative to request data from all channels of any specific Reactor at once. This way ensures that network latency is experienced only once per direction and Reactor. As with data messages, an unreliable network may make splitting requests, and subsequently, replies, into multiple messages in favor of smaller size preferable. Data requests MUST always be replied to immediately, but MAY be combined with other messages if such messages are about to be sent already. Waiting for additional requests is not allowed. An implementation MAY allow the user to set a maximum message length that best suits their network.

Information messages

An information message includes anything that are not channel data, like data types or GID assignments. They are therefore not merely "informational", as the name might suggest. Information messages usually appear during initialization of the Nexus, and therefore their impact on network performance is small during normal operation. They do not require acknowledgement, but may be combined with requests. They must be sent immediately, waiting for more information to be requested is not allowed.

Information requests

This includes anything that is neither a data request nor a periodic status inquiry, like requests for data types, Safe Values or Reactor type. Being the main cause of information messages, they also appear mostly during Nexus initialization, rendering their influence on network performance small. They may be combined with other information or data messages, but replies still MUST be sent immediately. Therefore, it is advisable to send all information requests for any specific Reactor within a single message to reduce the number of messages that have to be sent. However, if the network is very unreliable, a higher number of smaller messages may be preferable even for this kind of messages.

Lost nodes

Whenever a node does not receive replies (like periodic status inquiries) after 40 attempts, then the node must be considered as "lost". This will happen if the node has failed, is turned off or disconnected from the network. This applies to both Reactors and PSI Masters.

Timeouts

Whenever a Reactor does not receive data messages from the PSI Master, it MUST switch all channels to Safe Values. The time used for the timeout MUST be chosen according to the potential harm caused by the most dangerous channel of the Reactor, but SHOULD NOT be more than one second in any case. For dimmers, for example, this timeout is not critical, as they might cause fire only after some minutes, if at all. A LASER unit, especially if it does audience scanning, is much more critical, so the timeout time must follow local safety regulations unless other measures are put into place.
If a single channel doesn't receive values anymore, then it, too, must switch to it's Safe Value just like an entire Reactor. This also applies to channels using the Safety Sequence, that MUST be cancelled upon timeout.

Like any other channel, a channel using Safety Sequence times out if it does not receive new values. However, contrary to any other channel, this status is normal for an inactive channel using Safety Sequence: data for such channels needs to only be sent when the channel's value is supposed to change to or remain as anything possibly non-safe. Upon timeout, such a channel MUST therefore immediately cease operation and assure safety if at all possible. Only correctly received and matching parts of the Safety Sequence keep it from timing out.

If a PSI Master doesn't receive values from an Input or InOut channel or Reactor anymore, then it SHOULD set all channels that are controlled partially or in full by those channels to use Safe Values, especially if they use the Safety Sequence. In any case, use of Input or InOut channels to control safety-critical channels is not recommended, as values may be kept longer than they are valid.

4.  Discussion

As described in Section 3.1 (System-independent representation (transfer syntax)), a device independent data representation had to be chosen. The following options have been considered:

• ASN.1 (Abstract Syntax Notation, version 1) uses explicit typing. Each element is prepended with an unique identifier and length information, enabling the receiver to correctly receive and decode any message without knowledge of it's structure. The disadvantage is the additional overhead for transferring these metadata, which in case of the length may be up to 127 bytes (though usually one byte suffices for the length). The flexibility gained through explicitly encoding the type and length of each datum sent is not useful for PSI, because PSI defines all messages and their structure, meaning that if both sender and receiver conform to the PSI specification, an implicitly typed representation is perfectly sufficient. In fact, ASN.1 is mostly used for formal specifications instead of actual communication due to reasons like these. While ASN.1 offers data types of any size, down to a mere 8 bit, through the explicit typing and length bytes, this becomes at least three bytes when encoded as single data.
  
  
• CDR (Common Data Representation, [CORBA‑CDR] (Object Management Group, OMG., “Common Object Request Broker Architecture (CORBA) Specification, Version 3.1 - Part 2: CORBA Interoperability, section 9.3: CDR Transfer Syntax,” January 2008.)) is used in CORBA and defines a single format for every primitive type, but two endian representations that can be chosen by the sender. Therefore, every receiver must be prepared to byte-swap if needed. CDR offers data types down to 8 bit like ASN.1, without the overhead of explicit typing and length fields wherever possible, but requires data types to start at an aligned position matching their own size, with a few exceptions. This means overhead of a couple of bytes if a large data type follows smaller types that sum up to less than the alignment restriction of the larger type.
  
  However, some implementations, like the one of The Ace Orb ([TAO] (Schmidt, D., “The Ace Orb Home Page,” June 2007.)), already offer the option to turn off alignment restrictions. This unaligned CDR therefore is like XDR but with data types down to 8 bit in all cases.
  
  
• NDR (Network Data Representation) does not specify a single format definition. Instead it allows the sender to choose one and uses a Format Label to convey the type selected. Every receiver therefore has to provide all possible conversions, placing a significant burden on the implementor. Because in PSI every node is both sender and receiver, this burden would be placed on every single implementation. While this offers a speed advantage when similar hardware is communicating, PSI will encounter a highly heterogeneous environment, negating that advantage for the most part. NDR offers data types down to 8 bit like ASN.1, without the overhead of explicit typing and length fields wherever possible, but requires alignment of data types to their own size. Especially for the 8 bit Word types, this means overhead of a couple of bytes each.
  
  
• XDR (eXternal Data Representation, [RFC4506] (Eisler, M., “XDR: External Data Representation Standard,” May 2006.)) is used in a number of over-the-wire protocols. Like in ASN.1, the data are transferred as big endian. Contrary to ASN.1, XDR uses implicit typing and length like NDR, meaning that the receiver has to know beforehand which data type to expect next, but also resulting in less message and processing overhead. Contrary to NDR, it specifies a single format for all data types, so only one set of converters needs to be created. However, XDR requires 32 bit as smallest block size, so an 8 bit datum occupies 32 bit when encoded. This clearly wastes space, but can speed up encoding and decoding and ease implementation.

Since PSI uses a well-defined message format, lack of explicit typing and length information poses no problem, so the overhead created by ASN.1 would not be justified. While 8 bit data are the most common size at this time, ASN.1's advantage in this respect would be nearly negated because PSI uses explicit channel numbers. Because they are interspaced with the data values, every 8 bit datum would have to be encoded as three bytes, so PSI would not actually benefit much from the smaller data types. While PSI could have been designed to create contiguous ranges by, for example, specifying the channel numbers in one array and the data values in another, the added complexity would make PSI less easy to use, and a part of the flexibility aimed for would be lost. Additionally, this would only be feasible when the Nexus consists of Reactors that actually have a significant number of similar channels each, since otherwise separate messages or at least separate sentences would be required, with each containing the entire ASN.1 overhead.

CDR would, minus the alignment restrictions, clearly be the optimum choice regarding transfer efficiency, which is comparatively important, while the option of selecting endianness may or may not be viewed as a benefit. NDR is very similar, but it's cumbersome requirement of providing converters for a number of different representations per data type hinders easy development and creates larger probability for errors. ASN.1 may use Packed Encoding Rules (PER), but still would be less efficient than CDR, requires use of a special toolchain and PER is rather complex, creating even more possibility for error. XDR has the advantage of easy usability, topped up with availability: the runtime libraries required for ASN.1, CDR and NDR may be a portability issue, while XDR's comparatively simple implementations are present and readily available on platforms using libc and even µClibc. However, the presence of BSD-style open-source CDR implementations, along with the proliferation of CORBA implementations furthering CDR implementations on all major platforms means that adaptation issues are comparatively low.

In the end, unaligned CDR was chosen above the other options: it is available and significantly more bandwidth efficient than XDR. The bandwidth available from a typical, switched 100 Mbit network will suffice for a Nexus larger than two DMX universes even without use of any optimizations like Groups and Clusters, while maintaining a higher refresh rate even under the worst-case assumption that every single channel has an associated Reactor and thus requires an entire message for a single value. Similarly, 10 Gbit Ethernet will support a Nexus larger than 200 DMX universes, still at a higher refresh rate, over a single cable.

Of course, even without explicit typing done by the transfer syntax, a minimum of meta data still needs to be sent to identify the messages and their contents for flexibility. This results in a hybrid approach, which is somewhat similar to what is discussed in section 5 (6) of [RFC4506] (Eisler, M., “XDR: External Data Representation Standard,” May 2006.). It still allows for (comparatively) efficient utilization of the media while not sacrificing flexibility to any significant degree.

5.  Refresh rate

When running across a fully switched, 100 base T Ethernet, the PSI can achieve a worst-case refresh rate of over 110 Hz for 1000 channels. Compared to DMX this is about doubling both refresh rate and channel count at the same time. By using higher-level protocol functions like Groups, this can be increased to over 450 Hz with Groups of 20 Reactors each.

The assumption for these values is that every Reactor only contains exactly one channel, so that the entire overhead, including that of lower levels, is required for every single channel. If, under the same conditions as above, every Reactor hosts merely two channels, the refresh rate increases to over 850 Hz. Even though Input Reactors require two messages per refresh, assuming 100 Input Reactors of ten channels each, in Groups of five Reactors, refresh rates of over 700 Hz can be reached. Of course, instead of increasing the refresh rate, more channels can be added to the Nexus. With Gigabit Ethernet or faster being the norm, these figures increase appropriately.

Also, this comparison equates a DMX channel to a PSI channel, so the data and channel number size is 8 bits. If larger data sizes are used, then the comparative performance rises accordingly, but in reality most channels will use 8 bit resolution. Likewise, if larger channel numbers are used, then performance drops accordingly.
A large part of the loss of raw data efficiency compared to DMX is of course due to the protocol overhead that provides PSI with it's vastly increased flexibility, and to the lower layers (UDP, IP and Ethernet), that account for 66 bytes of overhead by themselves. Even so, a single 100 Mbit link suffices for most installations.

6.  Routing Considerations

The reliance on broadcasts for specific functions of this protocol implies that care must be taken when routers are employed. In normal environments, the presence of routers is neither required nor recommended, but appropriate addressing schemes may allow broadcasts to cross the routers. See [RFC1812] (Baker, F., “Requirements for IP Version 4 Routers,” June 1995.) and [RFC2644] (Senie, D., “Changing the Default for Directed Broadcasts in Routers,” August 1999.) for details. Reactors MUST be configurable to use unicast only, necessitating the entering of a fixed PSI Master address. The PSI Master should be designated the highest IP address in the given subnet by default, allowing Reactors to meaningfully default the unicast discovery target to this address without reconfiguration.
Additionally, nodes MAY implement the use of directed broadcasts for discovery, specifying a set of subnets that discovery messages are to be broadcast to.

7.  Security Considerations

This protocol is intended for use in a tightly controlled environment without publicly accessible connection points only. If a connection to external networks like the Internet exists, it is expected to be protected by appropriate measures like firewalling, that block all PSI related traffic in both directions. This assumption is reasonable given that the control wiring (rigging) is done outside of the audience's reach, and it may be assumed that the personnel with access to the rigging is reasonably trusted. Likewise, operations using this type of installation have no need for external connectivity on the production devices.
Similarly, the environment described is a low-risk target that would neither yield sensitive information when penetrated nor allow significant damage to be dealt when taken over or sabotaged.
Additionally, implementing security measures, even as simple as plain text passwords, would impair quick installation, exchanging of devices and automatic discovery, that are central features of the PSI. The cost of adding such security measures would therefore be unacceptably high compared to the rather low risk of attacks.
Therefore, no provisions are made for confidentiality, authentication or other security measures. Blind Denial of Service attacks are possible by sending Discovery messages of the FM type. If the attacker can send these messages as broadcast, one such message causes every Reactor to keep sending Discovery messages, and this will both multiply and keep going because the attacker will not acknowledge any such message. Additionally, an attacker might overflow the Reactor's list of PSI Masters by posing as more PSI Masters than it can hold. This requires reception of the Discovery messages however, so due to the routing restrictions listed in Section 6 (Routing Considerations) the attacker would need to be situated on the same network.
Should this protocol be required to span an unsafe network, existing security technologies like IPSec MUST be used. Depending on the actual security technology employed, the routing considerations in Section 6 (Routing Considerations) may or may not apply in this case.

While the Safety Sequence capability of the PSI requires an attacker to be able to receive messages in order to maliciously activate the devices, is not intended to safeguard against attacks; it is meant to protect against transmission errors, hardware or software failure, and misconfiguration. It is further noted that current controllers for this application may, in theory, be compromised as well, though the general availability of compatible networking devices, that is one of the strengths of this protocol, and the expected proliferation of unobserved connection points creates a much easier target than the comparatively direct wiring of a conventional controller.

8.  IANA Considerations

This protocol requires the assignment of two port numbers: one for the default PSI Master discovery reception port and one for the Reactor discovery reception port. Two distinct ports are necessary so that both PSI Master and Reactor may run as separate applications on the same IP address.

The ports assigned must be above 1023 so that the PSI does not require special privileges to be used. While only UDP is used for this protocol, the corresponding TCP ports might be assigned as well so that the management / control protocol that may be created at a later date will not require a new assignment. Currently, ports 7911 (short name: PSI/Reactor) and 4919 (short name: PSI/Master) are used.

9.  References

9.1. Normative References

9.2. Informative References

Author's Address