| Internet Engineering Task Force | A. Charny |
| Internet-Draft | Cisco Systems |
| Intended status: Experimental Protocol | F.Q. Huang |
| Expires: April 25, 2012 | Huawei Technologies |
| G. Karagiannis | |
| U. Twente | |
| M. Menth | |
| University of Tuebingen | |
| T. Taylor, Ed. | |
| Huawei Technologies | |
| October 23, 2011 |
PCN Boundary Node Behaviour for the Controlled Load (CL) Mode of Operation
draft-ietf-pcn-cl-edge-behaviour-10
Pre-congestion notification (PCN) is a means for protecting the quality of service for inelastic traffic admitted to a Diffserv domain. The overall PCN architecture is described in RFC 5559. This memo is one of a series describing possible boundary node behaviours for a PCN-domain. The behaviour described here is that for a form of measurement-based load control using three PCN marking states, not-marked, threshold-marked, and excess-traffic-marked. This behaviour is known informally as the Controlled Load (CL) PCN-boundary-node behaviour.
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 April 25, 2012.
Copyright (c) 2011 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.
The objective of Pre-Congestion Notification (PCN) is to protect the quality of service (QoS) of inelastic flows within a Diffserv domain, in a simple, scalable, and robust fashion. Two mechanisms are used: admission control, to decide whether to admit or block a new flow request, and (in abnormal circumstances) flow termination to decide whether to terminate some of the existing flows. To achieve this, the overall rate of PCN-traffic is metered on every link in the PCN-domain, and PCN-packets are appropriately marked when certain configured rates are exceeded. These configured rates are below the rate of the link thus providing notification to PCN-boundary-nodes about incipient overloads before any congestion occurs (hence the "pre" part of "pre-congestion notification"). The level of marking allows decisions to be made about whether to admit or terminate PCN-flows. For more details see [RFC5559].
Section 3 of this document specifies a detailed set of algorithms and procedures used to implement the PCN mechanisms for the CL mode of operation. Since the algorithms depend on specific metering and marking behaviour at the interior nodes, it is also necessary to specify the assumptions made about PCN-interior-node behaviour (Section 2). Finally, because PCN uses DSCP values to carry its markings, a specification of PCN-boundary-node behaviour MUST include the per domain behaviour (PDB) template specified in [RFC3086], filled out with the appropriate content (Section 4).
[RFC EDITOR'S NOTE: you may choose to delete the following paragraph and the "[CL-specific]" tags throughout this document when publishing it, since they are present primarily to aid reviewers. RFCyyyy is the published version of draft-ietf-pcn-sm-edge-behaviour.]
A companion document [RFCyyyy] specifies the Single Marking (SM) PCN-boundary-node behaviour. This document and [RFCyyyy] have a great deal of text in common. To simplify the task of the reader, the text in the present document that is specific to the CL PCN-boundary-node behaviour is preceded by the phrase: "[CL-specific]". A similar distinction for SM-specific text is made in [RFCyyyy].
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].
This document uses the following terms defined in Section 2 of [RFC5559]:
It also uses the terms PCN-traffic and PCN-packet, for which the definition is repeated from [RFC5559] because of their importance to the understanding of the text that follows:
This document uses the following terms from [RFC5670]:
To complete the list of borrowed terms, this document reuses the following terms and abbreviations defined in Section 3 of [RFC5696]:
This document defines the following additional terms:
This section describes the assumed behaviour for PCN-interior-nodes in the PCN-domain. The CL mode of operation assumes that:
This section describes the behaviour of the PCN-ingress-node, PCN-egress-node, and the Decision Point (which MAY be collocated with the PCN-ingress-node).
The PCN-egress-node collects the rates of not-marked, [CL-specific] threshold-marked, and excess-traffic-marked PCN-traffic for each ingress-egress-aggregate and reports them to the Decision Point. [CL-specific] It MAY also identify and report PCN-flows that have experienced excess-traffic-marking. For a detailed description, see Section 3.2.
The PCN-ingress-node enforces flow admission and termination decisions. It also reports the rate of PCN-traffic sent to a given ingress-egress-aggregate when requested by the Decision Point. For details, see Section 3.4.
Finally, the Decision Point makes flow admission decisions and selects flows to terminate based on the information provided by the PCN-ingress-node and PCN-egress-node for a given ingress-egress-aggregate. For details, see Section 3.3.
The PCN-egress-node MUST meter the PCN-traffic it receives in order to calculate the following rates for each ingress-egress-aggregate passing through it. These rates SHOULD be calculated at the end of each measurement period based on the PCN-traffic observed during that measurement period. The duration of a measurement period is equal to the configurable value T-meas.
Informative note: metering the PCN-traffic continuously and using equal-length measurement intervals minimizes the statistical variance introduced by the measurement process itself. On the other hand, the operation of PCN is not affected if the starting and ending times of the measurement intervals for different ingress-egress-aggregates are different.
[CL-specific] As a configurable option, the PCN-egress-node MAY record flow identifiers of the PCN-flows for which excess-traffic-marked packets have been observed during this measurement interval. If this set is large (e.g., more than 20 flows), the PCN-egress-node MAY record only the most recently excess-traffic-marked PCN-flow identifiers rather than the complete set.
Unless the report suppression option described in Section 3.2.3 is activated, the PCN-egress-node MUST report the latest values of NM-rate, [CL-specific] ThM-rate, and ETM-rate to the Decision Point each time that it calculates them.
[CL-specific] If the PCN-egress-node recorded a set of flow identifiers of PCN-flows for which excess-traffic-marking was observed in the most recent measurement interval, then it MUST also include these identifiers in the report.
Report suppression MUST be provided as a configurable option, along with two configurable parameters, the CLE-reporting-threshold and the maximum report suppression interval T-maxsuppress. The default value of the CLE-reporting-threshold is zero. The CLE-reporting-threshold MUST NOT exceed the CLE-limit configured at the Decision Point. For further information on T-maxsuppress see Section 3.5.
If the report suppression option is enabled, the PCN-egress-node MUST apply the following procedure to decide whether to send a report to the Decision Point, rather than sending a report automatically at the end of each measurement interval.
if any PCN-traffic was observed, or CLE = 0 if all the rates are zero.
Each report sent to the Decision Point when report suppression has been activated MUST contain the values of NM-rate, [CL-specific] ThM-rate, ETM-rate, and CLE that were calculated for the most recent measurement interval. [CL-specific] If the PCN-egress-node recorded a set of flow identifiers of PCN-flows for which excess-traffic-marking was observed in the most recent measurement interval, then it MUST also include these identifiers in the report.
The above procedure ensures that at least one report is sent per interval (T-maxsuppress + T-meas). This demonstrates to the Decision Point that both the PCN-egress-node and the communication path between that node and the Decision Point are in operation.
Operators can choose to use PCN procedures just for flow admission, or just for flow termination, or for both. A compliant Decision Point MUST implement both mechanisms, but configurable options MUST be provided to activate or deactivate PCN-based flow admission and flow termination independently of each other at a given Decision Point.
If PCN-based flow termination is enabled but PCN-based flow admission is not, flow termination operates as specified in this document.
The Decision Point determines the PCN-admission-state for a given ingress-egress-aggregate each time it receives a report from the egress node. It makes this determination on the basis of the congestion level estimate (CLE). If the CLE is provided in the egress node report, the Decision Point SHOULD use the reported value. If the CLE was not provided in the report, the Decision Point MUST calculate it based on the other values provided in the report, using the formula:
if any PCN-traffic was observed, or CLE = 0 if all the rates are zero.
The Decision Point MUST compare the reported or calculated CLE to a configurable value, the CLE-limit. If the CLE is less than the CLE-limit, the PCN-admission-state for that aggregate MUST be set to "admit"; otherwise it MUST be set to "block".
If the PCN-admission-state for a given ingress-egress-aggregate is "admit", the Decision Point SHOULD allow new flows to be admitted to that aggregate. If the PCN-admission-state for a given ingress-egress-aggregate is "block", the Decision Point SHOULD NOT allow new flows to be admitted to that aggregate. These actions MAY be modified by policy in specific cases, but such policy intervention risks defeating the purpose of using PCN.
[CL-specific] When the report from the PCN-egress-node includes a non-zero value of the ETM-rate for some ingress-egress-aggregate, the Decision Point MUST request the PCN-ingress-node to provide an estimate of the rate (PCN-sent-rate) at which the PCN-ingress-node is receiving PCN-traffic that is destined for the given ingress-egress-aggregate. Section 3.3.3 for a discussion of appropriate actions if the Decision Point fails to receive a timely response to its request for the PCN-sent-rate.
The Decision Point MUST then wait, for both the requested rate from the PCN-ingress-node and the next report from the PCN-egress-node for the ingress-egress-aggregate concerned. If this next egress node report also includes a non-zero value for the ETM-rate, the Decision Point MUST determine the amount of PCN-traffic to terminate using the following steps:
for the latest reported interval.
where PCN-sent-rate is the value provided by the PCN-ingress-node.
See
If the difference calculated in the second step is positive, the Decision Point SHOULD select PCN-flows to terminate, until it determines that the PCN-traffic admission rate will no longer be greater than the estimated sustainable aggregate rate. If the Decision Point knows the bandwidth required by individual PCN-flows (e.g., from resource signalling used to establish the flows), it MAY choose to complete its selection of PCN-flows to terminate in a single round of decisions.
Alternatively, the Decision Point MAY spread flow termination over multiple rounds to avoid over-termination. If this is done, it is RECOMMENDED that enough time elapse between successive rounds of termination to allow the effects of previous rounds to be reflected in the measurements upon which the termination decisions are based. (See [IEEE-Satoh] and sections 4.2 and 4.3 of [MeLe10].)
In general, the selection of flows for termination MAY be guided by policy. [CL-specific] If the egress node has supplied a list of identifiers of PCN-flows that experienced excess-traffic-marking (Section 3.2), the Decision Point SHOULD first consider terminating PCN-flows in that list.
The Decision Point SHOULD start a timer t-recvFail when it receives a report from the PCN-egress-node. t-recvFail is reset each time a new report is received from the PCN-egress-node. t-recvFail expires if it reaches the value T-fail. T-fail is calculated according to the following logic:
If timer t-recvFail expires for a given PCN-egress-node, the Decision Point SHOULD notify management. A Decision Point collocated with a PCN-ingress-node SHOULD cease to admit PCN-flows to the ingress-egress-aggregate associated with the given PCN-egress-node, until it again receives a report from that node. A centralized Decision Point MAY cease to admit PCN-flows to all ingress-egress-aggregates destined to the PCN-egress-node concerned, until it again receives a report from that node.
A centralized Decision Point SHOULD start a timer t-sndFail when it sends a request for the estimated value of PCN-sent-rate to a given PCN-ingress-node. If the Decision Point fails to receive a response from the PCN-ingress-node before t-sndFail reaches the configurable value T-crit, the Decision Point SHOULD repeat the request but MAY also use ETM-rate as an estimate of the amount of traffic to be terminated in place of the quantity Section 3.3.2. Because this will over-estimate the amount of traffic to be terminated due to dropping of PCN-packets by interior nodes, the Decision Point SHOULD use multiple rounds of termination under these circumstances. If the second request to the PCN-ingress-node also fails, the Decision Point SHOULD notify management.
specified in
See Section 3.5 for suggested values of the configurable durations T-crit and T-maxsuppress.
The PCN-ingress-node MUST provide the estimated current rate of PCN-traffic received at that node and destined for a given ingress-egress-aggregate in octets per second (the PCN-sent-rate) when the Decision Point requests it. The way this rate estimate is derived is a matter of implementation.
Here is a summary of the timers used in the procedures just described:
The timers just described depend on three configurable durations, T-meas, T-maxsuppress, and T-crit. The recommendations given below for the values of these durations are all related to the intended PCN reaction time of 1 to 3 seconds. However, they are based on judgement rather than operational experience or mathematical derivation.
The value of T-meas is RECOMMENDED to be of the order of 100 to 500 ms to provide a reasonable tradeoff between demands on network resources (PCN-egress-node and Decision Point processing, network bandwidth) and the time taken to react to impending congestion.
The value of T-maxsuppress is RECOMMENDED to be on the order of 3 to 6 seconds, for similar reasons to those for the choice of T-meas.
The value of T-crit SHOULD NOT be less than 3 * T-meas. Otherwise it could cause too many management notifications due to transient conditions in the PCN-egress-node or along the signalling path. A reasonable upper bound on T-crit is in the order of 3 seconds.
This section provides the specification required by [RFC3086] for a per-domain behaviour.
This section quotes [RFC5559].
The PCN CL boundary node behaviour specified in this document is applicable to inelastic traffic (particularly video and voice) where quality of service for admitted flows is protected primarily by admission control at the ingress to the domain.
In exceptional circumstances (e.g., due to rerouting as a result of network failures) already-admitted flows MAY be terminated to protect the quality of service of the remaining flows. [CL-specific] The performance results in, e.g., [MeLe10], indicate that the CL boundary node behaviour provides better service outcomes under such circumstances than the SM boundary node behaviour described in [RFCyyyy], because CL is less likely to terminate PCN-flows unnecessarily.
[RFC EDITOR'S NOTE: please replace RFCyyyy above by the reference to the published version of draft-ietf-pcn-sm-edge-behaviour.]
This section paraphrases the applicable portions of Sections 3.6 and 4.2 of [RFC5559].
Packets at the ingress to the domain are classified as either PCN or non-PCN. Non-PCN packets MAY share the network with PCN packets within the domain. Because the encoding specified in [RFC5696] and used in this document requires the use of the ECN fields, PCN-ingress-nodes MUST prevent ECN-capable traffic that uses the same DSCP as PCN from entering the PCN-domain directly. The PCN-ingress-node can accomplish this in three ways. The choice between these depends on local policy.
PCN packets are further classified as belonging or not belonging to an admitted flow. PCN packets not belonging to an admitted flow are dropped. Packets belonging to an admitted flow are policed to ensure that they adhere to the rate or flowspec that was negotiated during flow admission.
The PCN CL boundary node behaviour is a metering and marking behaviour rather than a scheduling behaviour. As a result, while the encoding uses a single DSCP value, that value MAY vary from one deployment to another. The PCN working group suggests using admission control for the following service classes (defined in [RFC4594]): [RFC5696].
For a fuller discussion, see Section A.1 of Appendix A of
The purpose of this per-domain behaviour is to achieve low loss and jitter for the target class of traffic. The design requirement for PCN was that recovery from overloads through the use of flow termination SHOULD happen within 1-3 seconds. PCN probably performs better than that.
The set of parameters that needs to be configured at each PCN-node and at the Decision Point is described in Section 5.1.
It is assumed that a specific portion of link capacity has been reserved for PCN-traffic.
The PCN CL behaviour MAY be used to carry real-time traffic, particularly voice and video.
The PCN CL per-domain behaviour can interfere with the use of end-to-end ECN due to reuse of ECN bits for PCN marking. See Appendix B of [RFC5696] for details.
Please see the security considerations in [RFC5559] as well as those in [RFC2474] and [RFC2475].
Deployment of the PCN Controlled Load edge behaviour requires the following steps:
The first set of decisions affects the operation of the network as a whole. To begin with, the operator needs to make basic design decisions such as whether the Decision Point is centralized or collocated with the PCN-ingress-nodes, and whether per-flow and aggregate resource signalling as described in [I-D.tsvwg-rsvp-pcn] is deployed in the network. After that, the operator needs to decide:
The following parameters affect the operation of PCN itself. The operator needs to choose:
Each PCN-ingress-node needs filters to classify incoming PCN packets according to ingress-egress-aggregate, both to satisfy Decision Point requests for sent traffic rates and, if applicable, to support encapsulation. If PCN packets are being tunneled to the PCN-egress-nodes (encapsulation), the PCN-ingress-node also needs the address of the PCN-egress-node for each ingress-egress-aggregate.
Each PCN-egress-node needs filters to classify incoming PCN packets by ingress-egress-aggregate, in order to gather measurements on a per-aggregate basis. The PCN-egress-node also needs to know the address of the Decision Point to which it sends reports for each ingress-egress-aggregate.
If [I-D.tsvwg-rsvp-pcn] is deployed and the Decision Points are collocated with the PCN-ingress-nodes, this information can be built up dynamically from the contents of the end-to-end RSVP signalling and does not have to be pre-configured. Otherwise the filters have to be derived from the routing tables in use in the domain, and the address of the peer at the other end of each ingress-egress-aggregate has to be tabulated for each PCN-edge-node.
A centralized Decision Point needs to have the address of the PCN-ingress-node corresponding to each ingress-egress-aggregate. Security considerations require that information also be prepared for a centralized Decision Point and each PCN-edge-node to allow them to authenticate each other.
Turning to link-specific parameters, the operator needs to derive values for the PCN-supportable-rate and [CL-specific] PCN-admissible-rate on each link in the network. The first two paragraphs of Section 5.2.2 of [RFC5559] discuss how these values may be derived. (For "PCN-excess-rate" in [RFC5559] read "PCN-supportable-rate", and [CL-specific] for "PCN-threshold-rate" read "PCN-admissible-rate".)
As discussed in the previous two sections, every PCN node needs to be provisioned with a number of parameters and policies relating to its behaviour in processing incoming packets. The Diffserv MIB [RFC3289] can be useful for this purpose, although it needs to be extended in some cases. This MIB covers packet classification, metering, counting, policing and dropping, and marking. The required extensions specifically include objects for re-marking the ECN field at the PCN-ingress-node and an extension to the classifiers to include the ECN field at PCN-interior and PCN-egress-nodes. In addition, the MIB has to be extended to include a potential encapsulation action following re-classification by ingress-egress-aggregate. Finally, new metering algorithms may need to be defined at the PCN-interior-nodes to handle threshold-marking and packet-size-independent excess-traffic-marking.
Values for the PCN-supportable-rate and [CL-specific] PCN-admissible-rate on each link on a node appear as metering parameters. Operators should take note of the need to deploy meters of a given type (threshold or excess-traffic) either on the ingress side or the egress of each interior link, but not both (Appendix B.2 of [RFC5670].
The following additional information has to be configured by other means (e.g., additional MIBs, NETCONF models).
At the PCN-egress-node:
At the Decision Point:
Depending on the testing strategy, it may be necessary to install the new configuration data in stages. This is discussed further below.
It is certainly not within the scope of this document to advise on testing strategy, which operators undoubtedly have well in hand. Quite possibly an operator will prefer an incremental approach to activation and testing. Implementing the PCN marking scheme at PCN-ingress-nodes, corresponding scheduling behaviour in downstream nodes, and re-marking at the PCN-egress-nodes is a large enough step in itself to require thorough testing before going further.
Testing will probably involve the injection of packets at individual nodes and tracking of how the node processes them. This work can make use of the counter capabilities included in the Diffserv MIB. The application of these capabilities to the management of PCN is discussed in the next section.
This section focuses on the use of event logging and the use of counters supported by the Diffserv MIB [RFC3289] for the various monitoring tasks involved in management of a PCN network.
It is anticipated that event logging using SYSLOG [RFC5424] will be needed for fault management and potentially for capacity management. Implementations MUST be capable of generating logs for the following events:
All of these logs are generated by the Decision Point. There is a strong likelihood in the first and third cases that the events are correlated with network failures at a lower level. This has implications for how often specific event types should be reported, so as not to contribute unnecessarily to log buffer overflow. Recommendations on this topic follow for each event report type.
Section 3.3.3 describes the circumstances under which the Decision Point may determine that it has lost contact, either with a PCN-ingress-node or a PCN-egress-node, due to failure to receive an expected report. Loss of contact with a PCN-ingress-node is a case primarily applicable when the Decision Point is in a separate node. However, implementations MAY implement logging in the collocated case if the implementation is such that non-response to a request from the Decision Point function can occasionally occur due to processor load or other reasons.
The log reporting the loss of contact with a PCN-egress-node MUST include the following content:
The following values are also RECOMMENDED for the indicated fields in this log, subject to local practice:
If contact is not regained with a PCN-egress-node in a reasonable period of time (say, one minute), the log SHOULD be repeated, this time with a PRI value of 113, implying a Severity value of "Alert: action must be taken immediately". The reasoning is that by this time, any more general conditions should have been cleared, and the problem lies specifically with the PCN-egress-node concerned and the PCN application in particular.
Whenever a loss-of-contact log is generated for a PCN-egress-node, a log indicating recovery SHOULD be generated when the Decision Point next receives a report from the node concerned. The log SHOULD have the same content as just described for the loss-of-contact log, with the following differences:
Section 3.3.2 describes the process whereby the Decision Point decides that flow termination is required for a given ingress-egress-aggregate, calculates how much flow to terminate, and selects flows for termination. This section describes a log that SHOULD be generated each time such an event occurs. (In the case where termination occurs in multiple rounds, one log SHOULD be generated per round.) The log may be useful in fault management, to indicate the service impact of a fault occuring in a lower-level subsystem. In the absence of network failures, it may also be used as an indication of an urgent need to review capacity utilization along the path of the ingress-egress-aggregate concerned.
The log reporting a flow termination event MUST include the following content:
The following values are also RECOMMENDED for the indicated fields in this log, subject to local practice:
The Diffserv MIB [RFC3289] allows for the provision of counters along the various possible processing paths associated with an interface and flow direction. It is RECOMMENDED that the PCN-nodes be instrumented as described below. It is assumed that the cumulative counts so obtained will be collected periodically for use in debugging, fault management, and capacity management.
PCN-ingress-nodes SHOULD provide the following counts for each ingress-egress-aggregate. Since the Diffserv MIB installs counters by interface and direction, aggregation of counts over multiple interfaces may be necessary to obtain total counts by ingress-egress-aggregate. It is expected that such aggregation will be performed by a central system rather than at the PCN-ingress-node.
PCN-interior-nodes SHOULD provide the following counts for each interface, noting that a given packet MUST NOT be counted more than once as it passes through the node:
PCN-egress-nodes SHOULD provide the following counts for each ingress-egress-aggregate. As with the PCN-ingress-node, so with the PCN-egress-node it is expected that any necessary aggregation over multiple interfaces will be done by a central system.
The following continuously cumulative counters SHOULD be provided as indicated, but require new MIBs to be defined. If the Decision Point is not collocated with the PCN-ingress-node, the latter SHOULD provide a count of the number of requests for PCN-sent-rate received from the Decision Point and the number of responses returned to the Decision Point. The PCN-egress-node SHOULD provide a count of the number of reports sent to each Decision Point. Each Decision Point SHOULD provide the following:
[RFC5559] provides a general description of the security considerations for PCN. This memo introduces no new considerations.
This memo includes no request to IANA.
The content of this memo bears a family resemblance to [ID.briscoe-CL]. The authors of that document were Bob Briscoe, Philip Eardley, and Dave Songhurst of BT, Anna Charny and Francois Le Faucheur of Cisco, Jozef Babiarz, Kwok Ho Chan, and Stephen Dudley of Nortel, Giorgios Karagiannis of U. Twente and Ericsson, and Attila Bader and Lars Westberg of Ericsson.
Ruediger Geib, Philip Eardley, and Bob Briscoe have helped to shape the present document with their comments. Toby Moncaster gave a careful review to get it into shape for Working Group Last Call.
Amongst the authors, Michael Menth deserves special mention for his constant and careful attention to both the technical content of this document and the manner in which it was expressed.
Finally, David Harrington's careful AD review resulted not only in necessary changes throughout the document, but also the addition of the operations and management considerations (Section 5).