| RMCAT WG | V. Singh |
| Internet-Draft | J. Ott |
| Intended status: Informational | Aalto University |
| Expires: April 23, 2014 | October 20, 2013 |
Evaluating Congestion Control for Interactive Real-time Media
draft-singh-rmcat-cc-eval-04
The Real-time Transport Protocol (RTP) is used to transmit media in telephony and video conferencing applications. This document describes the guidelines to evaluate new congestion control algorithms for interactive point-to-point real-time media.
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 23, 2014.
Copyright (c) 2013 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.
This memo describes the guidelines to help with evaluating new congestion control algorithms for interactive point-to-point real time media. The requirements for the congestion control algorithm are outlined in [I-D.jesup-rmcat-reqs]). This document builds upon previous work at the IETF: Specifying New Congestion Control Algorithms [RFC5033] and Metrics for the Evaluation of Congestion Control Algorithms [RFC5166].
The guidelines proposed in the document are intended to help prevent a congestion collapse, promote fair capacity usage and optimize the media flow's throughput. Furthermore, the proposed algorithms are expected to operate within the envelope of the circuit breakers defined in [I-D.ietf-avtcore-rtp-circuit-breakers].
This document only provides broad-level criteria for evaluating a new congestion control algorithm and the working group should expect a thorough scientific study to make its decision. The results of the evaluation are not expected to be included within the internet-draft but should be cited in the document.
The terminology defined in RTP [RFC3550], RTP Profile for Audio and Video Conferences with Minimal Control [RFC3551], RTCP Extended Report (XR) [RFC3611], Extended RTP Profile for RTCP-based Feedback (RTP/AVPF) [RFC4585] and Support for Reduced-Size RTCP [RFC5506] apply.
[RFC5166] describes the basic metrics for congestion control. Metrics that are of interest for interactive multimedia are:
Each experiment is expected to log every incoming and outgoing packet (the RTP logging format is described in Section 3.1). The logging can be done inside the application or at the endpoints using pcap (packet capture, e.g., tcpdump, wireshark). The following are calculated based on the information in the packet logs:
[Open issue (1): The "unfairness" test is (measured at 1s intervals):
1. Do not trigger the circuit breaker.
2. Over 3 times or less than 1/3 times the throughput for an RMCAT media stream compared to identical RMCAT streams competing on a bottleneck, for a case when the competing streams have similar RTTs.
3. Over 3 times delay compared to RTT measurements performed before starting the RMCAT flow or for the case when competing with identical RMCAT streams having similar RTTs.
]
[Open issue (2): Possibly using Jain-fairness index.]
Convergence time: the time taken to reach a stable rate at startup, after the available link capacity changes, or when new flows get added to the bottleneck link.
Bandwidth Utilization, defined as ratio of the instantaneous sending rate to the instantaneous bottleneck capacity. This metric is useful when an RMCAT flow is by itself or competing with similar cross-traffic.
From the logs the statistical measures (min, max, mean, standard deviation and variance) for the whole duration or any specific part of the session can be calculated. Also the metrics (sending rate, receiver rate, goodput, latency) can be visualized in graphs as variation over time, the measurements in the plot are at 1 second intervals. Additionally, from the logs it is possible to plot the histogram or CDF of packet delay.
The log file is tab or comma separated containing the following details:
Send or receive timestamp (unix)
RTP payload type
SSRC
RTP sequence no
RTP timestamp
marker bit
payload size
If the congestion control implements, retransmissions or FEC, the evaluation should report both packet loss (before applying error-resilience) and residual packet loss (after applying error-resilience).
A congestion control algorithm should be tested in simulation or a testbed environment, and the experiments should be repeated multiple times to infer statistical significance. The following guidelines are considered for evaluation:
The congestion control algorithm is expected to take an action, such as reducing the sending rate, when it detects congestion. Typically, it should intervene before the circuit breaker [I-D.ietf-avtcore-rtp-circuit-breakers] is engaged.
Does the congestion control propose any changes to (or diverge from) the circuit breaker conditions defined in [I-D.ietf-avtcore-rtp-circuit-breakers].
The congestion control should be assessed for its stability when the path characteristics do not change over time. Changing the media encoding rate estimate too often or by too much may adversely affect the application layer performance.
The congestion control algorithm should be assessed with different types of media behavior, i.e., the media should contain idle and data-limited periods. For example, periods of silence for audio, varying amount of motion for video, or bursty nature of I-frames.
The evaluation may be done in two stages. In the first stage, the endpoint generates traffic at the rate calculated by the congestion controller. In the second stage, real codecs or models of video codecs are used to mimic application-limited data periods and varying video frame sizes.
The congestion control algorithm should be assessed with different start-rates. The main reason is to observe the behavior of the congestion control in different evaluation scenarios, such as when competing with varying amount of cross-traffic or how quickly does the congestion control algorithm achieve a stable sending rate.
[Editor's note: requires a robust definition for unfriendliness and convergence time.]
The congestion control algorithm should be assessed in heterogeneous environments, containing both wired and wireless paths. Examples of wireless access technologies are: 802.11, GPRS, HSPA, or LTE. One of the main challenges of the wireless environments for the congestion control algorithm is to distinguish between congestion induced loss and transmission (bit-error corruption) loss. Congestion control algorithms may incorrectly identify transmission loss as congestion loss and reduce the media encoding rate by too much, which may cause oscillatory behavior and deteriorate the users' quality of experience. Furthermore, packet loss may induce additional delay in networks with wireless paths due to link-layer retransmissions.
The congestion control algorithm should be evaluated for a range of path characteristics such as, different end-to-end capacity and latency, varying amount of cross traffic on a bottleneck link and a router's queue length. For the moment, only DropTail queues are used. However, if new Active Queue Management (AQM) schemes become available, the performance of the congestion control algorithm should be again evaluated.
In an experiment, if the media only flows in a single direction, the feedback path should also be tested with varying amounts of impairments.
The main motivation for the previous and current criteria is to identify situations in which the proposed congestion control is less performant.
The congestion control algorithm should be able to handle changes in end-to-end capacity and latency. Latency may change due to route updates, link failures, handovers etc. In mobile environment the end-to-end capacity may vary due to the interference, fading, handovers, etc. In wired networks the end-to-end capacity may vary due to changes in resource reservation.
The congestion control algorithm should be evaluated when competing with other RTP flows using the same or another candidate congestion control algorithm. The proposal should highlight the bottleneck capacity share of each RTP flow.
[Editor's note: If we define Unfriendliness then that criteria should be applied here.]
The congestion control algorithm should be evaluated when competing with standard TCP. Short TCP flows may be considered as transient events and the RTP flow may give way to the short TCP flow to complete quickly. However, long-lived TCP flows may starve out the RTP flow depending on router queue length.
The proposal should also measure the impact on varied number of cross-traffic sources, i.e., few and many competing flows, or mixing various amounts of TCP and similar cross-traffic.
The congestion control algorithm should indicate if any protocol extensions are required to implement it and should carefully describe the impact of the extension.
[Editor's Note: If needed, a minimum evaluation criteria can be based on the above guidelines or defined tests/scenarios.]
An evaluation scenario is created from a list of network, link and flow characteristics. The example parameters discussed in the following subsections are meant to aid in creating evaluation scenarios and do not describe an evaluation scenario. The scenario discussed in Appendix B takes into account all these parameters.
The network scenario describes the types of flows sharing the common bottleneck with a single RMCAT flow, they are:
Figure 1 shows an example evaluation topology, where S1..Sn are traffic sources, these sources are either RMCAT or a mixture of traffic flows listed above. R1..Rn are the corresponding receivers. A and B are routers that can be configured to introduce impairments. Access links are in between the sender/receiver and the router, while the bottleneck link is between the Routers A and B.
+---+ Access Access +---+
|S1 |======= \ / =======|R1 |
+---+ link \\ // link +---+
\\ //
+---+ +-----+ Bottleneck +-----+ +---+
|S2 |=======| A |------------------------------>| B |=======|R2 |
+---+ | |<------------------------------| | +---+
+-----+ Link +-----+
(...) // \\ (...)
// \\
+---+ // \\ +---+
|Sn |====== / \ ======|Rn |
+---+ +---+
Figure 1: Simple Topology
[Open Issue: Discuss more complex topologies]
The media senders and receivers are typically connected to the bottleneck link, common access links are:
[Open issue: point to a reference containing parameters or traces to model WLAN and 3G/LTE.]
A real-world network typically consists of a mixture of links, the most important aspect is to identify the location of the bottleneck link. The bottleneck link can move from one node to another depending on the amount of cross-traffic or due to the varying link capacity. The design of the experiments should take this into account. In the simplest case the access link may not be the bottleneck link but an intermediate node.
The bottleneck link carries multiple flows, these flows may be other RMCAT flows or other types of cross-traffic. The experiments should dimension the bottleneck link based on the number of flows and the expected behavior. For example, if 5 media flows are expected to share the bottleneck link equally, the bottleneck link is set to 5 times the desired transmission rate.
If the experiment carries only media in one direction, then the upstream (sender to receiver) bottleneck link carries media packets while the downstream (receiver to sender) bottleneck carries the feedback packets. The bottleneck link parameters discussed in this section apply only to a single direction, hence the bottleneck link in the reverse direction can choose the same or have different parameters.
The link latency corresponds to the propagation delay of the link, i.e., the time it takes for a packet to traverse the bottleneck link, it does not include queuing delay. In an experiment with several links the experiment should describe if the links add latency or not. It is possible for experiments to have multiple hops with different link latencies. Experiments are expected to verify that the congestion control is able to work in challenging situations, for example over trans-continental and/or satellite links. The experiment should pick link latency values from the following:
Similarly, to model lossy links, the experiments can choose one of the following loss rates, the fractional loss is the ratio of packets lost and packets sent.
These fractional losses can be generated using traces, Gilbert-Elliot model, randomly (uncorrelated) loss.
The router queue length is measured as the time taken to drain the FIFO queue, they are:
However, the size of the queue is typically measured in bytes or packets and to convert the queue length measured in seconds to queue length in bytes:
QueueSize (in bytes) = QueueSize (in sec) x Throughput (in bps)/8
The media sources can be modeled in two ways. In the first, the sources always have data to send, i.e., have no data limited intervals and are able to generate the media rate requested by the RMCAT congestion control algorithm. In the second, the traffic generator models the behavior of a media codec, mainly the burstiness (time-varying data produced by a video GOP).
At the beginning of the session, the media sources are configured to start at a given start rate, they are:
Long-lived TCP flows will download data throughout the session and are expected to have infinite amount of data to send or receive.]
[Open issue: short-lived/bursty TCP cross-traffic parameters are still TBD.
Congestion control algorithms are expected to be published as "Experimental" documents until they are shown to be safe to deploy. An algorithm published as a draft should be experimented in simulation, or a controlled environment (testbed) to show its applicability. Every congestion control algorithm should include a note describing the environments in which the algorithm is tested and safe to deploy. It is possible that an algorithm is not recommended for certain environments or perform sub-optimally for the user.
[Editor's Note: Should there be a distinction between "Informational" and "Experimental" drafts for congestion control algorithms in RMCAT. [RFC5033] describes Informational proposals as algorithms that are not safe for deployment but are proposals to experiment with in simulation/testbeds. While Experimental algorithms are ones that are deemed safe in some environments but require a more thorough evaluation (from the community).]
Security issues have not been discussed in this memo.
There are no IANA impacts in this memo.
The content and concepts within this document are a product of the discussion carried out in the Design Team.
Michael Ramalho provided the text for the scenario discussed in Appendix B.
Much of this document is derived from previous work on congestion control at the IETF.
The authors would like to thank Harald Alvestrand, Luca De Cicco, Wesley Eddy, Lars Eggert, Kevin Gross, Vinayak Hegde, Stefan Holmer, Randell Jesup, Piers O'Hanlon, Colin Perkins, Michael Ramalho, Zaheduzzaman Sarker, Timothy B. Terriberry, Michael Welzl, and Mo Zanaty for providing valuable feedback on earlier versions of this draft. Additionally, also thank the participants of the design team for their comments and discussion related to the evaluation criteria.
| [RFC5033] | Floyd, S. and M. Allman, "Specifying New Congestion Control Algorithms", BCP 133, RFC 5033, August 2007. |
| [RFC5166] | Floyd, S., "Metrics for the Evaluation of Congestion Control Mechanisms", RFC 5166, March 2008. |
| [RFC5681] | Allman, M., Paxson, V. and E. Blanton, "TCP Congestion Control", RFC 5681, September 2009. |
| [SA4-EVAL] | R1-081955, 3GPP., "LTE Link Level Throughput Data for SA4 Evaluation Framework", 3GPP R1-081955, 5 2008. |
| [SA4-LR] | S4-050560, 3GPP., "Error Patterns for MBMS Streaming over UTRAN and GERAN", 3GPP S4-050560, 5 2008. |
| [TCP-eval-suite] | Lachlan, A., Marcondes, C., Floyd, S., Dunn, L., Guillier, R., Gang, W., Eggert, L., Ha, S. and I. Rhee, "Towards a Common TCP Evaluation Suite", Proc. PFLDnet. 2008, August 2008. |
Application trade-off is yet to be defined. see RMCAT requirements [I-D.jesup-rmcat-reqs] document. Perhaps each experiment should define the application's expectation or trade-off.
No quality metric is defined for performance evaluation, it is currently an open issue. However, there is consensus that congestion control algorithm should be able to show that it is useful for interactive video by performing analysis using a real codec and video sequences.
The goal of the experiment discussed in this section is to initially take out as many unknowns from the scenario. Later experiments can define more complex environments, topologies and media behavior. This experiment evaluates the performance of the RMCAT sender competing with other similar RMCAT flows (running the same algorithm or other RMCAT proposals) on the bottleneck link. There are up to 20 RMCAT flows competing for capacity, but the media only flows in one direction, from senders (S1..S20) to receivers (R1..R20) and the feedback packets flow in the reverse direction.
Figure 2 shows the experiment setup and it has subtle differences compared to the simple topology in Figure 1. Groups of 10 receivers are connected to the bottleneck link through two different routers (Router C and D). The rationale for adding these additional routers is to create two delay legs, i.e., two groups of endpoints with different network latencies and measure the performance of the RMCAT congestion control algorithm. If fewer than 10 sources are initialized, all traffic flows experience the same delay because they share the same delay leg.
Router A has a single forward direction bottleneck link (i.e., the bottleneck capacity and delay constraints applies only to the media packets going from the sender to the receiver, the feedback packets are unaffected). Hence, the Round-Trip Time (RTT) is primarily composed of the bottleneck queue delay and any forward path (propagation) latency. The main reason for not applying any constraints on the return path is to provide the best-case performance scenario for the congestion control algorithm. In later experiments, it is possible to add similar capacity and delay constraints on the return path.
+---+
/ === |R1 |
+---+ +-----+ // +---+
|S1 |======= \ / =| C | //
+---+ \\ // +-----+ \\ (...)
\\ // \\
+---+ +-----+ Bottleneck +-----+ \\ +---+
|S2 |=======| A |-------------------->| B | \ ===|R10|
+---+ | |<--------------------| | +---+
+-----+ Link +-----+
(...) // \\ +---+
// \\ / === |R11|
+---+ // \\ +-----+ // +---+
|S20|====== / \ =| D |//
+---+ +-----+\\ (...)
\\
\\ +---+
\ ===|R20|
+---+
Figure 2: Self-fairness Evaluation Setup
Loss impairments are applied at Router C and Router D, but only to the feedback flows. If the losses are set to 0%, it represents a case where the return path is over-provisioned for all traffic. In later experiments the loss impairments can be added to the media path as well.
The media sources are configured to send infinite amount of data, i.e., the sources always have data to send and have no data limited intervals. Additionally, the media sources are always successful in generating the media rate requested by the RMCAT congestion control algorithm. In this experiment, we avoid the potentially complicated scenario of using media traffic generators that try to model the behavior of media codecs (mainly the burstiness).
The media source always generates at the rate requested by the congestion control and has infinite data to send. Furthermore, the media packet generator is subject to the following constraints:
The bottleneck link capacity is dimensioned such that each RMCAT flow in an ideal situation with perfectly equal capacity sharing for all the flows on the bottleneck obtains the following throughputs: 200 kbps, 800 kbps, 1.3 Mbps and 4 Mbps.
For example, experiments with five RMCAT flows with an 800 kbps/flow target rate should set the bottleneck link capacity to 4 Mbps.
The bottleneck link queue (Router A) is a simple FIFO queue having a buffer length corresponding to 70 ms, 500 ms or 2000 ms (defined in Section 6.4) of delay at the bottleneck link rate (i.e., actual buffer lengths in bytes are dependent on bottleneck link bandwidth).
Experiments run with 1, 3, 5, 10 and 20 RMCAT sources, they are outlined as follows:
These cases assess if there are any early or late-comer advantages or disadvantages for a particular algorithm and to see if any unfairness is reproducible or unpredictable.
[Open issue (A.1): which group does the early and late flow belong to?]
[Open issue (A.2): Start rate for the media flows]
Packet loss is created in the reverse path (affects only feedback packets). Cases of 0%, 1%, 5% and 10% are studied for the 1, 3, and 5 RMCAT flow experiments, losses are not applied to flows with 10 or 20 RMCAT flows.
[Editor's note: there has been little or no discussion on the below criteria, however, they are listed here for the sake of completeness.
No unfairness is observed, i.e., at steady state each flow attains a throughput between [ B/(3*N), (3*B)/N ], where B is the link bandwidth and N is the number of flows.
No flow experiences packet loss when queue length is set to 500 ms or greater.
All individual sources must be in their steady state within twenty LRTTs (where LRTT is defined as the RTT associated with the flow with the Largest RTT in the experiment). ]
The above scenario describes only RMCAT sources competing for capacity on the bottleneck link, however, future experiments can use different types of cross-traffic (as described in Section 6.1).
Currently, the forward path (carrying media packets) is characterized to add delay and a fixed bottleneck link capacity, in the future packet losses and capacity changes can be applied to mimic a wireless link layer (for e.g., WiFi, 3G, LTE). Additionally, only losses are applied to the reverse path (carrying feedback packets), later experiments can apply the same forward path (carrying media packets) impairments to the reverse path.
Note to the RFC-Editor: please remove this section prior to publication as an RFC.