| Internet-Draft | DTP | July 2021 |
| Cui, et al. | Expires 26 January 2022 | [Page] |
This document defines Deadline-aware Transport Protocol (DTP) to provide block-based deliver-before-deadline transmission. The intention of this memo is to describe a mechanism to fulfill unreliable transmission based on QUIC as well as how to enhance timeliness of data delivery.¶
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 https://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 26 January 2022.¶
Copyright (c) 2021 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 (https://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.¶
Many emerging applications have the deadline requirement for their data transmission. However, current transport layer protocol like TCP [RFC0793]and UDP [RFC0768]only provide primitive connection establishment and data sending service. This document proposes a new transport protocol atop QUIC [QUIC]to deliver application data before end-to-end deadline.¶
The keywords MUST, MUST NOT, REQUIRED, SHALL, SHALL NOT, SHOULD, SHOULD NOT, RECOMMENDED, NOT RECOMMENDED, MAY, and OPTIONAL, when they appear in this document, are to be interpreted as described in [RFC2119].¶
Many applications such as real-time media and online multiplayer gaming have requirements for their data to arrive before a certain time i.e., deadline. For example, the end-to-end delay of video conferencing system should be below human perception (about 100ms) to enable smooth interaction among participants. For Online multiplayer gaming, the server aggregates each player's actions every 60ms and distributes these information to other players so that each player's state can be kept in sync. Data missing the deadline is useless since it will be overwrote by the new data.¶
These real-time applications have following common features:¶
However, current transport layer protocols lack support for block-based deadline delivery. TCP guarantees reliability so it will waste network resource to transmit stale data and cause fresh data to miss its deadline. UDP is unreliable but it doesn't drop data according to deadline, all data have the same chance to be dropped indeed. QUIC makes several improvements and introduces Stream Prioritization [QUIC]to enhance application performance, but prioritization is not enough for enhancing timeliness.¶
Insufficiency of existing transport layer forces applications to design their own customized and complex mechanism to meet the deadline requirement. For example, the video bitrate auto-adjustment in most streaming applications. But this is a disruption to the Layered Internet Architecture, forcing applications to worry about network conditions.¶
This document proposes Deadline-aware Transport Protocol (DTP) to provide deliver-before-deadline transmission. DTP is implemented as an extension of QUIC (Refer to [Section 4]) because QUIC provides many useful features including full encryption, user space deployment, zero-RTT handshake and multiplexing without head-of-line blocking.¶
The key insight of DTP is that these real-time applications usually have multiple blocks (As shown in Figure 1 below) to be transferred simultaneously and these blocks have diverse impact on user experience(denoted as priority). For example, audio data is more important than video stream in video conferencing. Central region is more important than surrounding region in 360 degree video. Foreground object rendering is more important than the background scene in mobile VR offloading.¶
The priority difference among multiple blocks makes it possible to drop low priority data to improve timeliness of high priority data delivery, which can enhance the overall QoE if resources allocated to blocks are correctly prioritized. In this section, we describe the mechanism which enables DTP to leverage that insight.¶
DTP provides block-based data abstraction for application. Application MUST attach metadata along with the data block to facilitate the scheduling decision, those metadata include:¶
The sender side architecture is shown in Figure 1:¶
+-------------+
| |
| Application |
| |
+-------------+
|
|
V
+------------------------------------------------------------------------+
| Block 0 Block 1 Block n |
| +--------+----------+ +--------+----------+ +--------+----------+ |
| |Metadata|Data Block| |Metadata|Data Block| ... |Metadata|Data Block| |
| +--------+----------+ +--------+----------+ +--------+----------+ |
| |
| (Metadata includes Deadline and Priority) |
+------------------------------------------------------------------------+
|
|
v
+-------------+
| |
| Scheduler |
| |
+-------------+
|
|
v
+-------------+
| |
| Redundancy |
| |
+-------------+
|
|
v
+-------------+
| |
| Congestion |
| Control |
+-------------+
|
v
Figure 1: The Architecture of DTP¶
In receiver side, the transport layer will receive data and reassemble the block. The process is symmetric with the sender side. It first goes through packet parsing and redundancy processing module. Transport layer also keeps track of the deadline of each block. When receiver calls RECV function (Refer to [Section 5]), the transport layer returns the received in-ordered data to the application.¶
The scheduler will pick the blocks to send and drop stale blocks when the buffer is limited. This section describes the algorithm of DTP scheduler.¶
Scheduler of DTP takes into account many factors when picking blocks in sender buffer to send. The goal of the scheduler is to deliver as much as high priority data before the deadline and drop obsolete or low-priority blocks. To achieve this, the scheduler utilizes both bandwidth and RTT measurement provided by the congestion control module and the metadata of blocks provided by the application to estimate the block completion time. The scheduler will run each time ACK is received or the application pushes the data.¶
A simple algorithm which only considers priority cannot get optimal result in transmitting deadline-required data. Suppose the bandwidth reduces and the scheduler chooses not to send the low priority block. Then the bandwidth is restored. The data block with lower priority is closer to the deadline than the high priority block. If in this round the scheduler still chooses to send the high priority block, then the low priority block may miss the deadline next round and become useless. In some cases, the scheduler can choose to send a low priority block because it's more urgent. But it should do so without causing the high priority stream missing the deadline. This example reveals a fundamental conflict between the application specified priority and deadline implicated priority. DTP needs to take both priorities into consideration when scheduling blocks.¶
DTP will combine all these factors to calculate real priority of each block. Then the scheduler just picks the block with the highest real priority. Scheduler of DTP will calculate the block remaining transmission time and then compare it to the deadline. The closer to the deadline, the higher real priority. And higher application specified priority will also result in higher real priority. In this way, the scheduler can take both approaching deadline and application-specified priority into account. Blocks which are severely overdue can be dropped accordingly.¶
After the scheduler pick the block to send, the packetizer will break the block into packet streams. Those packet streams will go through the redundancy module. When the link is lossy and deadline is tight, retransmission will cause the block missing the deadline. Redundancy module has the ability of sending redundancy (like FEC Repair Symbols) along with the data that will help to recover the data packets (like FEC Source Symbols), this can avoid retransmission.¶
We use unencrypted DTP packets as input to Redundancy Module because the loss of a DTP packet exactly corresponds to the loss of one Redundancy Packet. And to perform the coding and decoding with packets of different sizes, some packets may need to be padded with PADDING Frame. The present design of Redundancy Module follows the FEC Framework specified in [arXiv:1809.04822]. Figure 2 illustrates this framework:¶
|
|
v
+-------------+
| |
| DTP |
| Scheduler |
| |
+------+------+
|
(1)|DTP Packets
+----------|-------------------------------------------+
| v |
| +-------+------+ +------------+ |
| | | (2)DTP Payload | | | DTP
| | Redundancy |------------------> | Redundancy | | Redundancy
| | Packtizing &|<------------------ | Scheme | | Module
| | Grouping | (2)Redundancy Data | | |
| +-------+------+ +------------+ |
| | |
+----------|-------------------------------------------+
(3)|Redundancy
|Packets
|
v
Figure 2: DTP Redundancy Module¶
Figure 2 above shows the mechanism of how the Deadline-aware Redundancy module works. (1) Redundancy Module first receives the unencrypted DTP packets from scheduler. (2) The Redundancy Scheme use DTP Payload (similar to FEC Repair Symbols) to generate Redundancy Data (similar to FEC Source Symbols). (3) Redundancy-protected DTP Packets and Redundancy Data will be packtized and grouped. Redundancy Packtizing and Grouping Part will generate FEC Payload INFO (Figure 5) and attach it to the DTP Packets and Redundancy Data, generating Redundancy Packets (a Redundancy Packet with the header shown in Figure 5). Once the protocol receives the Repair Symbols, they are sent to the receiver through the FEC Packets. At the receiver-side, the received Redundancy Packets can be processed immediately. The Redundancy Data is reconstituted from the Redundancy Packtizing and Grouping and passed to the underlying Redundancy Scheme to recover the lost DTP Packets.¶
Although Redundancy Module allows recovering lost packets without waiting for retransmissions, it consumes more bandwidth than a regular, non-Redundancy-protected transmission. In order to avoid spending additional bandwidth when it is not needed, design of Redundancy MUST allow defining which DTP packets should be considered as Redundancy Packets. Currently we use a F flag from DTP Packet Header to indicate whether a packet is Redundancy-protected or not. The format of header will be described in [Section 4.3] later.¶
The Redundancy Data generated in Redundancy module MUST be distinguished from application data payload. Redundancy Data should not be transferred to the application upon reception, they are indeed generated by and for the Redundancy Scheme used by the transport protocol. We use Redundancy Packet to transmit Redundancy Data [Section 4.3].¶
There are multiple Redundancy Scheme candidates. During the handshake process, a scheme will be negotiated for the DTP session, just like encryption scheme negotiation. Currently DTP specifically chooses Reed-Solomon FEC Scheme as described in [arXiv:1809.04822].¶
This document reuses the congestion control module defined in QUIC [QUIC]. Congestion control module is responsible to send packets, collects ACK and do packet loss detection. Then it will put the lost data back to the retransmission queue of each block. Congestion control module is also responsible to monitor the network status and report the network condition such as bandwidth and RTT to scheduler.¶
DTP is implemented as an extension of QUIC by mapping QUIC stream to DTP block one to one. In that way, DTP can reuse the QUIC stream cancellation mechanism to drop the stale block during transmission. And DTP can also utilize the max stream data size defined by QUIC to negotiate its max block size. Besides, the block id of DTP can also be mapped to QUIC stream id without breaking the QUIC stream id semantic.¶
DTP endpoints communicate by exchanging packets. And the payload of DTP packets, consists of a sequence of complete frames. As defined in [QUIC], each frame begins with a Frame Type, indicating its type, followed by additional type-dependent fields. Besides the many frame types defined in Section 12.4 of [QUIC], DTP introduces BLOCK_INFO Frame to support timeliness data transmission. And DTP also makes adjustment on QUIC ACK Frame. Another extension is introducing FEC packet to support FEC.¶
DTP adds a BLOCK_INFO frame (type=0x20) in the front of each block to inform scheduler of Block Size, Block Priority and Block Deadline. These parameters can be used to do block scheduling. The BLOCK_INFO frame is as follows:¶
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
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| Stream ID (i) ...
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| Block Size (i) ...
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| Block Priority (i) ...
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| Block Deadline (i) ...
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
Figure 3: BLOCK_INFO Frame Format¶
DTP add a new Time Stamp Parameter to QUIC ACK Frame. Timestamped ACK frames are sent by reveiver to inform senders of the time when the packet the peer is acknowledging is received and processed. ACK mechanism of DTP is almost the same with QUIC. The format of the Timestamped ACK frames is similar to that of the standard ACK Frames defined in section 19.3 of [QUIC]:¶
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
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| Largest Acknowledged (i) ...
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| Time Stamp (i) ...
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| ACK Delay (i) ...
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| ACK Range Count (i) ...
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| First ACK Range (i) ...
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| ACK Ranges (i) ...
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| [ECN Counts] ...
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
Figure 4: Timestamped ACK Frame Format¶
Using this time stamp parameter we can calculate whether the prior blocks transmitted missing deadline or not, and we can also calculate the block completion rate before deadline.¶
We use a F Flags in DTP Packet to distinguish which DTP packets is Redundancy-protected or not. Figure 5 shows the Redundancy Packet Format. If the flag is set, the Redundancy Group ID, m, n, index field is appended to the header. They are used by the Redundancy Scheme(Forward-Error-Correction) to identify the redundancy-protected data and communicate information about the encoding and decoding procedures to the receiver-side Redundancy Scheme.¶
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
+-+-+-+-+-+-+-+-+
|F| Flags(7) |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
...
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| Redundancy Group ID (i)(if F set) ...
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| m (i)(if F set) ...
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| n (i)(if F set) ...
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| index (i)(if F set) ...
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| Payload (i) ...
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
Figure 5: Redundancy Packet Format¶
DTP can provide deliver-before-deadline service for Block Based Real Time Applications. Applications like real-time media and online multiplayer gaming have deadline requirements for their data transimission. These application also tend to generate and process the data in block fashion, for example, video/audio encoder produces the encoded streams as a series of block (I,B,P frame or GOP). And these real-time applications usually have multiple blocks (As shown in Figure 1) to be transferred simultaneously. DTP can optimize the data transmission of these applications by scheduling which block to be sent first. And Redundancy Module of DTP can reduce retransmission delay.¶
DTP extends the send socket API to let application attach metadata along with the data block, and the API of DTP is structured as follows:¶
Send¶
Update¶
Retreat¶
Receive¶
on_dropped¶
on_delivered¶
All these functions mentioned above are running in asynchronous mode. An application can use various event driven framework to call those functions.¶
This document has no actions for IANA.¶
See the security considerations in [QUIC]and [QUIC-TLS]; the block-based data of DTP shares the same security properties as the data transmitted within a QUIC connection¶