C.Burmeister Internet Draft R.Hakenberg draft-burmeister-avt-rtcp-feedback-sim-00.txt A.Miyazaki Expires: April 2002 Matsushita J.Ott University of Bremen TZI N.Sato S.Fukunaga Oki November 2001 Extended RTP Profile for RTCP-based Feedback - Results of the Timing Rule Simulations - Status of this Memo This document is an Internet-Draft and is in full conformance with all provisions of Section 10 of RFC2026. Internet-Drafts are working documents of the Internet Engineering Task Force (IETF), its areas, and its working groups. Note that other groups may also distribute working documents as Internet- Drafts. 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." The list of current Internet-Drafts can be accessed at http://www.ietf.org/ietf/1id-abstracts.txt The list of Internet-Draft Shadow Directories can be accessed at http://www.ietf.org/shadow.html. Abstract This document describes the results we achieved when simulating the timing rules of the Extended RTP Profile for RTCP-based Feedback. Unicast and multicast topologies are considered as well as several protocol and environment configurations. The results show that the timing rules result in better performance regarding feedback delay and still preserve the well accepted RTP rules regarding allowed bit rates for control traffic. Burmeister et al. Expires May 2002 1 Extended RTP Profile for RTCP-based Feedback November 2001 - Results of the Timing Rule Simulations - Table of Contents Status of this Memo Abstract 1 Introduction 2 Conventions used in this document 3 Timing rules of the extended RTP profile for RTCP-based feedback 4 Simulation Environment 4.1 Network Simulator Version 2 4.2 RTP Agent 4.3 Scenarios 4.4 Topologies 5 RTCP Bit Rate Measurements 5.1 Unicast 5.2 Multicast 5.3 Summary of the RTCP bit rate measurements 6 Feedback Measurements 6.1 Unicast 6.2 Multicast 6.2.1 Shared Losses vs Distributed Losses 6.2.2 Sender vs. Receiver 7 Investigations on "k" 7.1 Feedback Suppression Performance 7.2 Loss Report Delay 7.3 Summary of "k" investigations 8 Investigations on "l" 8.1 Feedback Suppression Performance 8.2 Loss Report Delay 8.3 Summary of "l" investigations 9 Applications Using AVPF 9.1 NEWPRED Implementation in NS2 9.2 Simulation 9.2.1. Simulation A - Constant Packet Loss Rate 9.2.2. Simulation B - Packet Loss due to Congestion 9.3 Summary 10 Summary References Authors Addresses Burmeister et al. Expires May 2002 2 Extended RTP Profile for RTCP-based Feedback November 2001 - Results of the Timing Rule Simulations - 1 Introduction The Real-time Transport Protocol (RTP) is widely used for the transmission of real-time or near real-time media data over the Internet. While it was originally designed to work well for multicast groups in very large scales, its scope is not limited to that. More and more applications use RTP for small multicast groups (e.g. video conferences) or even unicast (e.g. media streaming applications). RTP comes together with its companion protocol Real-time Transport Control Protocol (RTCP), which is used to monitor the transmission of the media data and provide feedback of the reception quality. What is more it can be used for a loosely session control. Having the scope of large multicast groups in mind, the rules when to send feedback were much restricted to avoid feedback explosion or feedback related congestion in the network. RTP and RTCP have proven to work well in the Internet, especially in large multicast groups, which is shown by its tremendous usages today. However the applications that transmit the media data only to small multicast groups or unicast, may benefit from more frequent feedback. The source of the packets might be able to react to changes in the reception quality, which might be due to congestion in the network or other sudden changes. Possible reactions include sending rate adaptation according to a congestion control algorithm or the invocation of error resilience features for the media stream (e.g. retransmissions, reference picture selection, NEWPRED, etc.). As said before, more feedback would be needed to increase the reception quality, but RTP restricts the use of RTCP feedback very much. Hence it was decided to create a new extended RTP profile, which redefines some of the RTCP timing rules, but keeps most of the algorithms for RTP and RTCP, which have proven to work well. The new rules should scale from unicast to multicast, where unicast or small multicast applications have the most gain from it. A detailed description of the new profile and its timing rules can be found in [1]. This document investigates the new algorithms by the means of simulations. We show that the new timing rules scale and behave network friendly. Therefore we first describe roughly the key features of the new RTP profile, which are important for our simulations, in Section 3. After that we describe the environment that is used to conduct the simulations in Section 4. Section 5 describes simulation results that show the backwards compatibility to RTP and that the new profile is network friendly in terms of used bit rate for feedback and other control traffic. In Section 6 we show the benefit that applications could get from implementing the new profile. In Section 7 and 8 we show the merit for some special Burmeister et al. Expires May 2002 3 Extended RTP Profile for RTCP-based Feedback November 2001 - Results of the Timing Rule Simulations - parameters settings and finally in section 9 we show the performance gain we could get for a special application, namely NEWPRED in MPEG-4. 2 Conventions used in this document 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 RFC-2119. 3 Timing rules of the extended RTP profile for RTCP-based feedback As said above, RTP restricts the usage of RTCP feedback. The main rules that restrict the feedback are as follows: - RTCP messages are sent in compound packets, i.e. every RTCP packet contains at least one sender report (SR) or receiver report (RR) message and a source description (SDES) message. - The RTCP compound packets are sent in time intervals (T_rr), which is computed as a function of the average packet size, the number of senders and receivers in the group and the session bandwidth. (-> 5% of the session bandwidth is used for RTCP messages; this bandwidth is shared between all session members, where the senders might get more than the receivers.) - The minimum interval between two RTCP packets from the same source is 5 seconds. We see that these rules prevent feedback explosion and scale to very large multicast groups. However they do not allow timely feedback at all. While the second rule scales also to small groups or unicast (in this cases the interval might be as small as a few milliseconds), the third rule prevents the receivers from sending feedback in time. The timing rules to send RTCP feedback from the new RTP profile [1] consists of two key components. First the minimum interval of 5 seconds is abolished. Second, receivers get once during their (now quite small) RTCP interval the chance to send an RTCP packet "early", i.e. not according to the calculated interval, but virtually immediately. It is important to note that the RTCP interval calculation is still inherited from the original RTP specification. The specification and all the details of the extended timing rules can be found in [1]. We do not want to describe the algorithms here, but rather reference these from the original specification where needed. Therefore we use also the same variable names and abbreviations as in [1]. Burmeister et al. Expires May 2002 4 Extended RTP Profile for RTCP-based Feedback November 2001 - Results of the Timing Rule Simulations - 4 Simulation Environment This section describes the simulator that was used for the investigations and its key features. The extensions to the simulator, that were necessary are described roughly. 4.1 Network Simulator Version 2 The simulations were conducted using the network simulator version 2 (ns2). ns2 is an open source project, written in a combination of Tool Command Language (TCL) and C++. The scenarios are set-up using TCL. In the scripts it is possible to specify the topologies (nodes and links, bandwidths, queue sizes or error rates for links) and the parameters of the "agents", i.e. protocol configurations. The protocols itself are implemented in C++ in the agents, which are connected to the nodes. A detailed description of ns2 and a downloadable newest version can be found at [4]. 4.2 RTP Agent We implemented a new agent, based on RTP/RTCP. RTP packets are sent at a constant packet rate with the correct header sizes. RTCP packets are sent according to the timing rules of [2] and also its algorithms for group membership maintenance are implemented. Sender and receiver reports are sent and the senders use these reports to maintain a RTT estimation to the other group members, as it is described in [2]. Further we extended the agent to support the extended profile [1]. The use of the new timing rules can be turned on and off via parameter settings in TCL. 4.3 Scenarios The scenarios that are simulated are defined in TCL scripts. We set- up several different topologies, ranging from unicast with two session members to multicast with up to 25 session members. Depending on the used sending rates and the corresponding link bandwidths congestion losses may occur. In some scenarios, bit errors are inserted on certain links. We simulated groups with RTP/AVP agents, RTP/AVPF agents and mixed groups. The feedback messages are generally NACK messages as defined in [1] and are triggered by packet loss. Burmeister et al. Expires May 2002 5 Extended RTP Profile for RTCP-based Feedback November 2001 - Results of the Timing Rule Simulations - 4.4 Topologies Mainly four different topologies are simulated to show the key features of the extended profile. However for some specific simulations we used, different topologies, which is then indicated at the description of the simulation results. The main four topologies are named after the number of participating RTP agents, i.e. T-2, T-4, T-8 and T-16, where T-2 is a unicast scenario, T-4 contains four agents, etc. The figures below illustrate the main topologies. A5 A5 | A6 / | / / | /--A7 / |/ A2 A2-----A6 A2--A8 / / / A9 / / / / / / / /---A10 A1-----A2 A1-----A3 A1-----A3-----A7 A1------A3< \ \ \ \---A11 \ \ \ \ \ \ \ A12 A4 A4-----A8 A4--A13 |\ | \--A14 | \ | A15 A16 T-2 T-4 T-8 T-16 Figure 1: Simulated Topologies. 5 RTCP Bit Rate Measurements The new timing rules allow more frequent RTCP feedback for small multicast groups. In large groups the algorithm behaves similar to usual RTP. While it is generally good to have more frequent feedback it cannot be allowed at all to increase the bit rate used for RTCP above a fixed limit, i.e. 5% of the total RTP bandwidth according to RTP. This section shows that with the new timing rules we keep the 5% limit for all investigated scenarios, topologies and group sizes. What is more, we show that mixed groups, i.e. some members use AVP some use AVPF, can be allowed and that each session member behaves fair according to its corresponding specification. Burmeister et al. Expires May 2002 6 Extended RTP Profile for RTCP-based Feedback November 2001 - Results of the Timing Rule Simulations - 5.1 Unicast First we measured the RTCP bandwidth share in the unicast topology T-2. Even for a fixed topology and group size, there are several protocol parameters which are varied to simulate a large range of different scenarios. First we varied the RTP session bandwidth. For large session bandwidths, the allowed RTCP bit rate increases also and thus more RTCP packets can be sent. Second we changed the number of agents that are pure receivers or also senders. This has also some influence on the RTCP feedback, because on the one hand pure receivers do not have an RTT estimation and one the other hand they do not send sender reports. Third we varied the configurations of the agents in that sense that the agents may use the AVP or AVPF. Thereby it is possible that one agent uses AVP and the other AVPF in one RTP session. This is done to test the backwards compatibility. First we consider scenarios where no losses occur. In this case both RTP session members transmit the RTCP compound packets at regular intervals, calculated as T_rr, if they use the AVPF, and use the minimum interval of 5s if they implement the AVP. No early packets are sent, because the need to send feedback is not given. Still it is important to see that not more than 5% of the session bandwidth is used for RTCP and that AVP and AVPF members can co-exist without interference. The results can be found in table 1. | | | | | | Used RTCP Bit Rate | | Session | Send | Rec. | AVP | AVPF | (% of session bw) | |Bandwidth|Agents|Agents|Agents|Agents| A1 | A2 | sum | +---------+------+------+------+------+------+------+------+ | 2 Mbps | 1 | 2 | - | 1,2 | 2.42 | 2.56 | 4.98 | | 2 Mbps | 1,2 | - | - | 1,2 | 2.49 | 2.49 | 4.98 | | 2 Mbps | 1 | 2 | 1 | 1,2 | 0.01 | 2.49 | 2.50 | | 2 Mbps | 1,2 | - | 1 | 1,2 | 0.01 | 2.48 | 2.49 | | 2 Mbps | 1 | 2 | 1,2 | 1,2 | 0.01 | 0.01 | 0.02 | | 2 Mbps | 1,2 | - | 1,2 | 1,2 | 0.01 | 0.01 | 0.02 | |200 kbps | 1 | 2 | - | 1,2 | 2.42 | 2.56 | 4.98 | |200 kbps | 1,2 | - | - | 1,2 | 2.49 | 2.49 | 4.98 | |200 kbps | 1 | 2 | 1 | 1,2 | 0.06 | 2.49 | 2.55 | |200 kbps | 1,2 | - | 1 | 1,2 | 0.08 | 2.50 | 2.58 | |200 kbps | 1 | 2 | 1,2 | 1,2 | 0.06 | 0.06 | 0.12 | |200 kbps | 1,2 | - | 1,2 | 1,2 | 0.08 | 0.08 | 0.16 | | 20 kbps | 1 | 2 | - | 1,2 | 2.44 | 2.54 | 4.98 | | 20 kbps | 1,2 | - | - | 1,2 | 2.50 | 2.51 | 5.01 | | 20 kbps | 1 | 2 | 1 | 1,2 | 0.58 | 2.48 | 3.06 | | 20 kbps | 1,2 | - | 1 | 1,2 | 0.77 | 2.51 | 3.28 | | 20 kbps | 1 | 2 | 1,2 | 1,2 | 0.58 | 0.61 | 1.19 | | 20 kbps | 1,2 | - | 1,2 | 1,2 | 0.77 | 0.79 | 1.58 | Table 1: Unicast simulations without packet loss. Burmeister et al. Expires May 2002 7 Extended RTP Profile for RTCP-based Feedback November 2001 - Results of the Timing Rule Simulations - We can see that in configurations, where both Agents use the new timing rules each of them uses about 2.5% of the session bandwidth for RTP, which sums up to 5% of the session bandwidth for both. This is achieved regardless of the agent being a sender or a receiver. In the cases where Agent1 uses AVP and Agent2 AVPF, the total RTCP session bandwidth is decreased. This is due to the fact that Agent1 can send RTCP packets only with a minimum interval of 5 seconds. Thus only a small fraction of the session bandwidth is used for its RTCP packets. For a high bit rate session (session bandwidth = 2 Mbps) the fraction of the RTCP packets from Agent one is as small as 0.01%. For smaller session bandwidths the fraction increases, because the same amount of RTCP data is sent. The bandwidth share that is used by RTCP packets from Agent 2 is not different from what was used, when both Agents implemented the AVPF. Thus the interaction of AVP and AVPF agents is not problematic in these scenarios at all. In our second unicast experiment, we show that the allowed RTCP bandwidth share is not exceeded, even if packet loss occurs. We simulated a constant byte error rate (BYER) on the link. The byte errors are inserted randomly with a uniform distribution. Packets with byte errors are discarded on the link; hence the receiving agents will not see the loss immediately. The agents detect packet loss by a gap in the sequence number. When the agents detect a packet loss, they feel the need to send feedback. In unicast T_dither_max is always zero, hence an early packet can be sent immediately if allow_early is true. If the last packet was already an early one (i.e. allow_early = false), the feedback might be appended to the next regularly scheduled receiver report. The max_feedback_delay parameter (which we set to 1 second in our simulations) determines if that is allowed. The results are shown in table 2, where we can see that there is no difference in the RTCP bandwidth share, whether losses occur or not. This is what we expected, because even though the RTCP packet size grows and early packets are sent, the interval between the packets increases and thus the RTCP bandwidth stays the same. Only the RTCP bandwidth of the Agents that use the AVP increases slightly. This is because the interval between the packets is still 5 seconds, but the packet size increased because of the feedback that is appended. Burmeister et al. Expires May 2002 8 Extended RTP Profile for RTCP-based Feedback November 2001 - Results of the Timing Rule Simulations - | | | | | | Used RTCP Bit Rate | | Session | Send | Rec. | AVP | AVPF | (% of session bw) | |Bandwidth|Agents|Agents|Agents|Agents| A1 | A2 | sum | +---------+------+------+------+------+------+------+------+ | 2 Mbps | 1 | 2 | - | 1,2 | 2.42 | 2.56 | 4.98 | | 2 Mbps | 1,2 | - | - | 1,2 | 2.49 | 2.49 | 4.98 | | 2 Mbps | 1 | 2 | 1 | 1,2 | 0.01 | 2.49 | 2.50 | | 2 Mbps | 1,2 | - | 1 | 1,2 | 0.01 | 2.48 | 2.49 | | 2 Mbps | 1 | 2 | 1,2 | 1,2 | 0.01 | 0.02 | 0.03 | | 2 Mbps | 1,2 | - | 1,2 | 1,2 | 0.01 | 0.01 | 0.02 | |200 kbps | 1 | 2 | - | 1,2 | 2.42 | 2.56 | 4.98 | |200 kbps | 1,2 | - | - | 1,2 | 2.50 | 2.49 | 4.99 | |200 kbps | 1 | 2 | 1 | 1,2 | 0.06 | 2.50 | 2.56 | |200 kbps | 1,2 | - | 1 | 1,2 | 0.08 | 2.49 | 2.57 | |200 kbps | 1 | 2 | 1,2 | 1,2 | 0.06 | 0.07 | 0.13 | |200 kbps | 1,2 | - | 1,2 | 1,2 | 0.09 | 0.08 | 0.17 | | 20 kbps | 1 | 2 | - | 1,2 | 2.42 | 2.57 | 4.99 | | 20 kbps | 1,2 | - | - | 1,2 | 2.52 | 2.51 | 5.03 | | 20 kbps | 1 | 2 | 1 | 1,2 | 0.58 | 2.54 | 3.12 | | 20 kbps | 1,2 | - | 1 | 1,2 | 0.83 | 2.43 | 3.26 | | 20 kbps | 1 | 2 | 1,2 | 1,2 | 0.58 | 0.73 | 1.31 | | 20 kbps | 1,2 | - | 1,2 | 1,2 | 0.86 | 0.84 | 1.70 | Table 2: Unicast simulations with packet loss. 5.2 Multicast Next we investigated the RTCP bandwidth share in multicast scenarios, i.e. we simulated the topologies T-4, T-8 and T-16 and measured the fraction of the session bandwidth that was used for RTCP packets. Again we considered different situations and protocol configurations (e.g. with or without bit errors, groups with AVP and/or AVPF agents, etc.). For reasons of readability, we present only selected results. For a documentation of all results, see [5]. The simulations of the different topologies in scenarios, where no losses occur, neither through bit errors nor through congestion, show a similar behavior as the unicast scenarios. For all group sizes the maximum used RTCP bit rate share is 5.06% of the session bandwidth in a simulation of 16 session members in a low bit rate scenario (session bandwidth = 20kbps) with several senders. In all other scenarios without losses the used RTCP bit rate share is below that. Thus the requirement, that not more than 5% of the session bit rate should be used for RTCP is fulfilled in reasonable accuracy. Simulations, were bit errors are randomly inserted in RTP and RTCP packets and the corrupted packets are discarded, give the same results. The 5% rule is kept (at maximum 5.07% of the session bandwidth is used for RTCP). Burmeister et al. Expires May 2002 9 Extended RTP Profile for RTCP-based Feedback November 2001 - Results of the Timing Rule Simulations - Finally we conducted simulations, where we reduced the link bandwidth and thereby caused congestion related losses. These simulations are different from the previous bit error simulations, in that the losses occur more in bursts and are more correlated, also between different agents. The correlation and burstness of the packet loss is due to the queuing discipline in the routers we simulated; we used simple FIFO queues with a drop-tail strategy to handle congestion. Random Early Detection (RED) queues may enhance the performance, because the burstness of the packet loss might be reduced, however this is not subject of our investigations, but is left for future research. The delay between the agents, which also influence RTP and RTCP packets, is much more variable because of the added queuing delay. Still the used RTCP bit rate share does not increase beyond 5.09% of the session bandwidth. Thus also for these special cases the requirement is fulfilled. 5.3 Summary of the RTCP bit rate measurements We have shown that for unicast and reasonable multicast scenarios, feedback explosion does not happen. The requirement that at maximum 5% of the session bandwidth is used for RTCP is fulfilled for all investigated scenarios. 6 Feedback Measurements In this chapter we describe the results of feedback delay measurements, we conducted in the simulations. Therefore we use two metrics for measuring the performance of the algorithms, these are the mean "waiting time" (MWT) and the number of feedback that is sent, suppressed or not allowed. The waiting time is the time, measured at a certain agent, between the detection of a packet loss and the time when the corresponding feedback is sent. Assuming that the value of the feedback decreases with its delay, we think that the mean waiting time is a good metric to measure the performance gain we could get by using AVPF instead of AVP. The feedback an agent wants to send can be either sent or not sent. If it was not sent, this could be due to the feedback suppression, i.e. another receiver already sent the same feedback or because the feedback was not allowed, i.e. the max_feedback_delay was exceeded. We traced for every detected loss, if the agent sent the corresponding feedback or not and if not, why. The more feedback was not allowed, the worse the performance of the algorithm. Together with the waiting times, this gives us a good hint of the overall performance of the scheme. Burmeister et al. Expires May 2002 10 Extended RTP Profile for RTCP-based Feedback November 2001 - Results of the Timing Rule Simulations - 6.1 Unicast In the unicast case, the maximum dithering interval T_Dither_max is fixed and set to zero. This is due to the fact that it does not make sense for a unicast receiver to wait for other receivers if they have the same feedback to send. But still feedback can be delayed or might not be permitted to be sent at all. The dithering interval is a parameter for the early packets, but at maximum every second packet can be an early packet. The regularly scheduled packets are spaced according to T_rr, which depends in the unicast case mainly on the session bandwidth. Table 3 shows the mean waiting times (MWT) for some configurations of the unicast topology T-2. The number of feedback packets that are sent or discarded is listed also (feedback sent (sent) or feedback discarded (disc)). We do not list suppressed packets, because for the unicast case feedback suppression does not apply. In the simulations, agent 1 was a sender and agent 2 a pure receiver. We did not vary this, because the only difference in being a sender or pure receiver, is that the sender has an RTT estimation to the receivers. However the RTT estimation is used for the T_Dither_max calculations only in the multicast cases. | | | Feedback Statistics | | Session | | AVP | AVPF | |Bandwidth| PLR | sent |disc| MWT | sent |disc| MWT | +---------+-------+------+----+-------+------+----+-------+ | 2 Mbps | 0.001 | 781 | 0 | 2.604 | 756 | 0 | 0.015 | | 2 Mbps | 0.01 | 7480 | 0 | 2.591 | 7548 | 2 | 0.006 | | 2 Mbps | cong. | 25 | 0 | 2.557 | 1741 | 0 | 0.001 | | 20 kbps | 0.001 | 79 | 0 | 2.472 | 74 | 2 | 0.034 | | 20 kbps | 0.01 | 780 | 0 | 2.605 | 709 | 64 | 0.163 | | 20 kbps | cong. | 780 | 0 | 2.590 | 687 | 70 | 0.162 | Table 3: Feedback Statistics for the unicast simulations. From the table above we see that the mean waiting time can be decreased dramatically by using AVPF instead of AVP. While the waiting times for agents using AVP is always around 2.5 seconds (half the minimum interval) it can be decreased to a few ms for most of the AVPF configurations. In the cases of high session bandwidth normally all feedback is sent. This is because the packet size is quite large (1000byte) and thus per lost packet, more RTCP bandwidth is available. There are only very few exceptions, which are probably due to two packet losses within one RTCP interval, where the first loss was by chance sent quite early. In this case it might be possible that the second feedback is detected after the early packet was sent, but too early Burmeister et al. Expires May 2002 11 Extended RTP Profile for RTCP-based Feedback November 2001 - Results of the Timing Rule Simulations - to append it to the next regularly scheduled report, because of the limitation of the max_feedback_delay. This is different for the cases with a small session bandwidth. Here we have a small packet size (100byte) and thus many packets are transmitted, while the RTCP bandwidth share is quite low. T_rr is thus quite large. After an early packet was sent the time to the next regularly scheduled packet can be very high. We saw that in some cases the time was larger than than max_feedback_delay, because in these cases the feedback is not allowed to be sent at all. With a different setting of max_feedback_delay it is possible to have either more feedback that is not allowed and a decreased mean waiting time or more feedback that is sent but an increased waiting time. Thus the parameter should be set with care according to the application's needs. 6.2 Multicast In this section we describe some measurements of feedback statistics in the multicast simulations. We picked out certain characteristic and representative results. Therefore we considered the topology T- 16. Different scenarios and applications are simulated for this topology. The parameters of the different links are set as follows. The agents A2, A3 and A4 are connected to the middle node of the multicast tree, i.e. agent A1, via high bandwidth and low delay links. The other agents are connected to the nodes 2, 3 and 4 via different link characteristics. The agents connected to node 2 represent mobile users. They suffer in certain configurations from a certain byte error rate on their access links and the delays are quite high. The agents that are connected to node 3 have low bandwidth access links, but do not suffer from bit errors. The last agents, that are connected to node 4 have quite high bandwidth and quite low delay. 6.2.1 Shared Losses vs Distributed Losses In our first investigation, we wanted to see the influence the loss characteristic on the algorithm's performance, i.e. we wanted to investigate the cases where packet loss occurs for several users simultaneously or totally independently. Therefore we first define agent A1 to be the sender. In the shared-loss-case we insert a constant byte error rate on one of the middle links, i.e. the link between A1 and A2. In the case of distributed losses we inserted the same byte error rate on all links downstream of A2. This scenario is especially interesting, because of the feedback suppression algorithm. When all receivers share the same loss, it is only necessary for one of them to send the loss report. Hence if a member receives feedback with the same content that it has scheduled Burmeister et al. Expires May 2002 12 Extended RTP Profile for RTCP-based Feedback November 2001 - Results of the Timing Rule Simulations - to be sent, it suppresses the scheduled feedback. Of course this suppressed feedback does not contribute to the mean waiting times. So we expect reduced waiting times for shared losses, because the probability is high that one of the receivers can send the feedback more or less immediately. The results are shown in the following table. | | Feedback Statistics | | | Shared Losses | Distributed Losses | |Agent|sent|fbsp|disc|sum | MWT |sent|fbsp|disc|sum | MWT | +-----+----+----+----+----+-----+----+----+----+----+-----+ | A2 | 274| 351| 25| 650|0.267| -| -| -| -| -| | A5 | 231| 408| 11| 650|0.243| 619| 2| 32| 653|0.663| | A6 | 234| 407| 9| 650|0.235| 587| 2| 32| 621|0.701| | A7 | 223| 414| 13| 650|0.253| 594| 6| 41| 641|0.658| | A8 | 188| 443| 19| 650|0.235| 596| 1| 32| 629|0.677| Table 4: Feedback statistics for multicast simulations. Table 4 shows the feedback statistics for the simulation of a large group size. All 16 agents of topology T-16 joined the RTP session. However only agent A1 acts as an RTP sender, the other agents are pure receivers. Only 4 or 5 agents suffer from packet loss, i.e. A2, A5, A6, A7 and A8 for the case of shared losses and A5, A6, A7 and A8 in the case of distributed losses. Since the number of session members is the same for both cases, T_rr is also the same on the average. Still the mean waiting times are reduced by more than 50% in the case of shared losses. This proves our assumption that shared losses enhance the performance of the algorithm. The feedback suppression mechanism seems to be working quite fine. Even though some feedback is sent from different receivers (i.e. 1150 loss reports are sent in total and only 650 packets were lost, resulting in loss report being received on the average 1.8 times) most of the redundant feedback was suppressed. I.e. 2023 loss reports were suppressed from 3250 individual detected losses, which means that more than 60% of the feedback was actually suppressed. 6.2.2 Sender vs. Receiver RTP senders are able to maintain a RTT measurement to all receivers, which send receiver reports. This is done by the means of the ntp timestamp in the sender report and the repetition of this value together with the delay since last sender report value in the receiver report. However RTP session members that do not send RTP packets are not an RTP sender and thus do not send sender reports. Therefore pure receivers do not have an RTT measurement to the senders or other receivers. This fact is considered in AVPF, by giving two possibilities to calculate T_dither_max. Burmeister et al. Expires May 2002 13 Extended RTP Profile for RTCP-based Feedback November 2001 - Results of the Timing Rule Simulations - If the RTP member has an RTT measurement to the sender of the packet it wants to provide feedback to, it calculates T_dither_max = k * T_rtt/2 * members, with k = 1. Thus t_dither_max is increased with the number of session members and the RTT. The rational for RTT/2 is that the distance to the sender is a good measure how long to wait at maximum. Other receivers, who are more far away, i.e. have a larger RTT estimation, will detect the packets later and also the feedback from those would arrive later and hence have less value. Thus the nearest receivers get the chance first to send their feedback. Because of the larger distance of the other receivers to the sender, they will probably wait longer (probably, because of the randomness, i.e. we calculate T_dither_max, from which T_dither is picked randomly). While those are waiting, it is likely that they receive the feedback from the receivers that are nearer to the source. With this it is possible to find a good compromise between waiting time and feedback suppression. To let the algorithm scale to large group sizes, the number of session members is included. The number of members is the maximum number of receivers that shared the same loss. The more members are in the session, the higher is the probability that other receivers share the loss and thus the higher is the value of waiting longer, because the probability is increased that feedback suppression will work. If all receivers calculate the same T_dither_max ( i.e. have a similar RTT estimation) and pick a T_dither from this interval randomly with a uniform distribution, it is likely that one feedback is sent within the first RTT interval. In case the RTP session member does not have an RTT measurement, i.e. it is a pure receiver, is calculates T_dither_max = l * T_rr, with l = 0.5. The rational for this is that the receiver, if it has no RTT estimation, does not know at all how long it should wait for other receivers to send feedback. The feedback suppression algorithm would certainly fail, if the time is selected too short. However the waiting time is increased unnecessarily (and thus the value of the feedback is decreased!) in case the time is chosen too long. It would be good to find the optimum time (which is tried to be done with the RTT estimation), but it is not dangerous if the optimum time is not chosen. Decreased feedback value and a failure of the feedback suppression mechanism do not hurt the network stability. We have shown for the cases of distributed losses that the overall bandwidth constraints are kept in any case and thus we could only loose some performance by choosing the wrong time. A good measure for T_dither_max however is the RTCP interval T_rr. This value increases with the number of session members. Also we know that we can send feedback at least every T_rr. Thus increasing T_dither max beyond T_rr would certainly make no sense. So by choosing T_rr/2 we guarantee that at least sometimes (i.e. when a loss is detected in the first half of the interval between two regularly scheduled RTCP packets) we are allowed to send early packets. Because of the randomness of T_dither we still have a good chance to send the early packet in time. Burmeister et al. Expires May 2002 14 Extended RTP Profile for RTCP-based Feedback November 2001 - Results of the Timing Rule Simulations - Having said that, we assume that the RTP members who have an RTT measurement would perform better regarding the feedback suppression. We want to show that by simulating the same scenario of the previous section, but enabling all receivers that suffer from packet loss to maintain a RTT measurement. We do this by declaring the corresponding agents to RTP senders. However we do not send RTP packets from this agents, to be comparable to the previous results. The only difference to the previous simulations is that sender reports are sent, which enables the sender to maintain a RTT measurement. | | Feedback Statistics | | | Shared Losses | Distributed Losses | |Agent|sent|fbsp|disc|sum | MWT |sent|fbsp|disc|sum | MWT | +-----+----+----+----+----+-----+----+----+----+----+-----+ | A2 | 582| 43| 7| 632|0.100| -| -| -| -| -| | A5 | 70| 562| 0| 632|0.121| 644| 1| 1| 646|0.576| | A6 | 60| 572| 0| 632|0.114| 638| 5| 1| 644|0.575| | A7 | 73| 559| 0| 632|0.109| 607| 3| 1| 611|0.567| | A8 | 63| 569| 0| 632|0.108| 626| 3| 0| 629|0.589| Table 5: Feedback statistics for multicast simulations, where the agents that suffer from packet loss do have an RTT estimation to the sender. Table 5 shows the results of the simulations. As assumed, we see that the performance regarding the waiting time is increased significantly. In case of shared losses, the mean time is less than half of the mean waiting times of the receivers that do not have a RTT estimation. Also for the case of distributed losses, we see a slight gain in performance, however not as big as for the shared losses. But still we see that the calculation of T-dither_max, using the RTT estimation finds a better tradeoff between waiting time and feedback suppression. The waiting time is reduced and the feedback suppression increased where possible. Thus for both cases, whether feedback suppression is possible or not, the performance is increased. Feedback suppression in the case of shared losses is working much better with a RTT estimation. From 3160 individual detected losses only 848 loss reports are sent. 7 Investigations on "k" The parameter k in the formula how to calculate T_Dither_max if an RTT estimation is available has some influence of the performance of the algorithm. Thus we investigated the effect and tried to find an optimum value for k. Therefore we defined a sample scenarios and tried to find an optimum value for k. Burmeister et al. Expires May 2002 15 Extended RTP Profile for RTCP-based Feedback November 2001 - Results of the Timing Rule Simulations - We define three representative sample scenarios. We use the topology from the previous section. Most of the agents however contribute only little to the simulations, because we introduced an error rate only on the link between the sender A1 and the agent A2. The first scenario represents cases, where losses are shared between two agents. One agent is located upstream on the path between the other agent and the sender. Therefore agent A2 and agent A5 see the same losses, that are introduce on the link between the sender and agent A2. Agent A6, A7 and A8 do not join the RTP session. From the other agents only agents A3 and A9 join. Both agent A2 and A5 are declared as RTP senders, in order to have an RTT estimation to the sender A1. The second scenario represents also cases, where losses are shared between two agents, but this time the agents are located on different branches of the multicast tree. The delays to the sender are roughly of the same magnitude. Agent A5 and A6 share the same losses. Agents A3 and A9 join the RTP session, but are pure receivers and do not see any losses. Also in the third scenario, the losses re shared between two agents, A5 and A6. The same agents as in the second scenario are active. However the delays of the links are different. The delay of the link between agent A2 and A5 is reduced to 20ms and between A2 and A6 to 40ms. Thus the RTT estimations of agents A5 and A6 to the sender are reduced significantly. 7.1 Feedback Suppression Performance First we consider the fraction of feedback that the agent An suppresses (Feedback Suppression Rate). An is thereby the agent nearer to the source. The simulation results can be seen from Table 7. In general it can be seen that agent An suppresses more feedback if the differences between the delays to the source are smaller. This is reasonable, because the feedback from other receivers will be faster received in that case. It can also be seen that the feedback suppression rate increases with k. This is due to the fact that T_dither_max increases with k. Thus the agents will wait longer on the average before sending their feedback. By increasing the waiting time for all agents, the time were feedback suppression is possible at all is increased. Burmeister et al. Expires May 2002 16 Extended RTP Profile for RTCP-based Feedback November 2001 - Results of the Timing Rule Simulations - | | Feedback Suppression Rate | | k | Scen. 1 | Scen. 2 | Scen. 3 | +------+---------+---------+---------+ | 0.10 | 0.070 | 0.039 | 0.064 | | 0.25 | 0.068 | 0.063 | 0.065 | | 0.50 | 0.062 | 0.114 | 0.124 | | 0.75 | 0.047 | 0.172 | 0.129 | | 1.00 | 0.056 | 0.234 | 0.176 | | 1.25 | 0.056 | 0.282 | 0.233 | | 1.50 | 0.047 | 0.315 | 0.251 | | 1.75 | 0.040 | 0.331 | 0.245 | | 2.00 | 0.048 | 0.297 | 0.284 | | 3.00 | 0.047 | 0.347 | 0.330 | | 4.00 | 0.063 | 0.347 | 0.353 | Table 7: Fraction of feedback that was suppressed at agent An of the total number of feedback the agent wanted to send In Table 8 the results for the feedback suppression of agent Af are depicted. Again we see that the number of feedback suppressions increase with k. Only in scenario 1 the number is more or less constant. However by increasing the waiting times, the probability that the feedback is suppressed is decreased at agent Af. k=1 seems to be a threshold, where the feedback suppression does not change anymore significantly in the given scenarios. This is because for the given parameters, the early packets will not be sent any more, because the next regularly scheduled RTCP packet will we within the T_dither_max interval. | | Feedback Suppression Rate | | k | Scen. 1 | Scen. 2 | Scen. 3 | +------+---------+---------+---------+ | 0.10 | 0.736 | 0.064 | 0.071 | | 0.25 | 0.814 | 0.079 | 0.119 | | 0.50 | 0.859 | 0.162 | 0.239 | | 0.75 | 0.865 | 0.222 | 0.376 | | 1.00 | 0.844 | 0.290 | 0.401 | | 1.25 | 0.850 | 0.338 | 0.429 | | 1.50 | 0.849 | 0.316 | 0.473 | | 1.75 | 0.868 | 0.316 | 0.505 | | 2.00 | 0.843 | 0.376 | 0.487 | | 3.00 | 0.845 | 0.345 | 0.502 | | 4.00 | 0.820 | 0.345 | 0.493 | Table 8 Fraction of feedback that was suppressed at agent Af of the total number of feedback the agent wanted to send In Table 9 the ration of feedback suppression failures is illustrated. In general the observations from the figures above are summarized. The ratio of feedback failures decreases with an Burmeister et al. Expires May 2002 17 Extended RTP Profile for RTCP-based Feedback November 2001 - Results of the Timing Rule Simulations - increasing k for the scenarios 2 and 3. In scenario 1 the ratio is hardly influenced at all by k. The simulations show a kind of steady state at k larger two or three, where the rationale for this is that for very large k, T_dither_max becomes equal or more than T_rr and thus no early packets are send any more. The maximum dithering interval is for these cases limited by the next regularly scheduled RR. | |Feedback Suppr. Failure Rate | | k | Scen. 1 | Scen. 2 | Scen. 3 | +------+---------+---------+---------+ | 0.10 | 0.194 | 0.897 | 0.865 | | 0.25 | 0.117 | 0.858 | 0.816 | | 0.50 | 0.079 | 0.725 | 0.638 | | 0.75 | 0.088 | 0.606 | 0.495 | | 1.00 | 0.100 | 0.468 | 0.423 | | 1.25 | 0.094 | 0.381 | 0.338 | | 1.50 | 0.104 | 0.369 | 0.276 | | 1.75 | 0.092 | 0.353 | 0.250 | | 2.00 | 0.110 | 0.328 | 0.229 | | 3.00 | 0.108 | 0.308 | 0.169 | | 4.00 | 0.116 | 0.308 | 0.154 | Table 8: The ratio of feedback suppression failures. Summarizing, it can be said, that the feedback suppression performance is highly dependent on the topology, the parameters and configurations. In general a larger value for k increases the probability that the feedback suppression works, however the performance gain decreases with an increasing k. For a certain threshold, depending on the configuration and environment, an increasing k does not lead to any performance gain any more. 7.2 Loss Report Delay In this section we investigate the influence of the parameter k on the loss report delay. Therefore we measured for the three sample scenarios the mean loss report delay as seen by the sender, i.e. the sender calculates for every loss report, it receives for the first time the delay since the corresponding packet was sent. The results are depicted in Table 9. In general it can be said, that the loss report delay increases with k. This is only natural, because T_Dither_max is proportional to k. Thus the agents wait on the average longer to send their early packets. In cases of very large k values, the report delay does not increase significantly any more. In these cases nearly no early packets are sent, because the Burmeister et al. Expires May 2002 18 Extended RTP Profile for RTCP-based Feedback November 2001 - Results of the Timing Rule Simulations - next regularly scheduled packet is within the T_dither_max interval. The threshold of k, from which on the delay will not increase, is dependent on the RTT estimation. For increasing RTT values, the threshold decreases. We see that in scenario 1 the threshold lies between k=2 and k=3. For the scenarios with smaller RTT, the threshold is higher. Summarizing it can be said, that the report delay increases with an increasing k. From a certain threshold the increase is not significant, however this threshold is highly dependent on topology and environment parameters. | | Mean Loss Report Delay | | k | Scen. 1 | Scen. 2 | Scen. 3 | +------+---------+---------+---------+ | 0.10 | 0.128 | 0.282 | 0.431 | | 0.25 | 0.135 | 0.266 | 0.430 | | 0.50 | 0.150 | 0.264 | 0.497 | | 0.75 | 0.160 | 0.286 | 0.538 | | 1.00 | 0.194 | 0.305 | 0.613 | | 1.25 | 0.203 | 0.329 | 0.661 | | 1.50 | 0.208 | 0.363 | 0.690 | | 1.75 | 0.209 | 0.387 | 0.739 | | 2.00 | 0.242 | 0.412 | 0.764 | | 3.00 | 0.243 | 0.507 | 0.790 | | 4.00 | 0.287 | 0.568 | 0.790 | Table 9: The mean loss report delay, measured at the sender. 7.3 Summary of "k" investigations We have shown by simulations that the parameter k influence the feedback performance. While in general the feedback suppression performance increases with k, the report delay increases also. Hence we need to find a tradeoff, between the amount of feedback that is sent and the delay of the feedback, when it is received at the sender. Since we have shown that the performance curves for the feedback suppression as well as the report delay is highly variable for different topologies and environments, it is not possible to give an optimized parameter value for k. We think that k=1 is a compromise, which should be acceptable for most of our considered cases. At least we guarantee with k=1 that no feedback explosion will occur and thus keep the network stability untouched. Burmeister et al. Expires May 2002 19 Extended RTP Profile for RTCP-based Feedback November 2001 - Results of the Timing Rule Simulations - 8 Investigations on "l" In this section we want to investigate the influence of the parameter "l" from the T_Dither_max calculation in agents that do not have an RTT estimation to the sender. As we have done in the previous section for the parameter "k", we investigate the feedback suppression performance as well as the report delay for three sample scenarios. For simplicity we use the same scenarios as in the previous section, but this time the all agents beside agent A1 are pure RTP receivers. Thus these agents do not have an RTT estimation to the source. T_Dither_Max is calculated with the other formula, depending only on T_rr and l, which means that all agents should calculate roughly the same T_Dither_Max. 8.1 Feedback Suppression Performance The results for the feedback suppression rate of the agent Af that is more far away from the sender, are depicted in Table 10. In general it can be seen that the feedback suppression rate increases with an increasing l. However there is a threshold, depending on the environment, from which the additional gain is not significant any more. | | Feedback Suppression Rate | | l | Scen. 1 | Scen. 2 | Scen. 3 | +------+---------+---------+---------+ | 0.10 | 0.671 | 0.051 | 0.089 | | 0.25 | 0.582 | 0.060 | 0.210 | | 0.50 | 0.524 | 0.114 | 0.361 | | 0.75 | 0.523 | 0.180 | 0.370 | | 1.00 | 0.523 | 0.204 | 0.369 | | 1.25 | 0.506 | 0.187 | 0.372 | | 1.50 | 0.536 | 0.213 | 0.414 | | 1.75 | 0.526 | 0.215 | 0.424 | | 2.00 | 0.535 | 0.216 | 0.400 | | 3.00 | 0.522 | 0.220 | 0.405 | | 4.00 | 0.522 | 0.220 | 0.405 | Table 10: Fraction of feedback that was suppressed at agent An of the total number of feedback the agent wanted to send Burmeister et al. Expires May 2002 20 Extended RTP Profile for RTCP-based Feedback November 2001 - Results of the Timing Rule Simulations - Similar results can be seen for the agent that is nearer to the sender in Table 11. | | Feedback Suppression Rate | | l | Scen. 1 | Scen. 2 | Scen. 3 | +------+---------+---------+---------+ | 0.10 | 0.056 | 0.056 | 0.090 | | 0.25 | 0.063 | 0.055 | 0.166 | | 0.50 | 0.116 | 0.099 | 0.255 | | 0.75 | 0.141 | 0.141 | 0.312 | | 1.00 | 0.179 | 0.175 | 0.352 | | 1.25 | 0.206 | 0.176 | 0.361 | | 1.50 | 0.193 | 0.193 | 0.337 | | 1.75 | 0.197 | 0.204 | 0.341 | | 2.00 | 0.207 | 0.207 | 0.368 | | 3.00 | 0.196 | 0.203 | 0.359 | | 4.00 | 0.196 | 0.203 | 0.359 | Table 11: Fraction of feedback that was suppressed at agent An of the total number of feedback the agent wanted to send The rate of feedback suppression failure is depicted in Table 12. The trend that the additional performance increase is not significant from a certain threshold, depending on the environment is here as well visible. | |Feedback Suppr. Failure Rate | | l | Scen. 1 | Scen. 2 | Scen. 3 | +------+---------+---------+---------+ | 0.10 | 0.273 | 0.893 | 0.822 | | 0.25 | 0.355 | 0.885 | 0.624 | | 0.50 | 0.364 | 0.787 | 0.385 | | 0.75 | 0.334 | 0.679 | 0.318 | | 1.00 | 0.298 | 0.621 | 0.279 | | 1.25 | 0.289 | 0.637 | 0.267 | | 1.50 | 0.274 | 0.595 | 0.249 | | 1.75 | 0.274 | 0.580 | 0.235 | | 2.00 | 0.258 | 0.577 | 0.233 | | 3.00 | 0.282 | 0.577 | 0.236 | | 4.00 | 0.282 | 0.577 | 0.236 | Table 12: The ratio of feedback suppression failures. Summarizing the feedback suppression results it can be said that in general the feedback suppression performance increases with an increasing l. However from a certain threshold, depending on environment parameters such as propagation delays or session bandwidth, the additional increase is not significant anymore. Burmeister et al. Expires May 2002 21 Extended RTP Profile for RTCP-based Feedback November 2001 - Results of the Timing Rule Simulations - 8.2 Loss Report Delay In this section we show the results for the measured report delay during the simulations of the three sample scenarios. This measurement is a metric of the performance of the algorithms, because the value of the feedback for the sender typically decreases with the delay of its reception. The loss report delay is measured as the time at the sender between sending a packet and receiving the first corresponding loss report. | | Mean Loss Report Delay | | l | Scen. 1 | Scen. 2 | Scen. 3 | +------+---------+---------+---------+ | 0.10 | 0.124 | 0.282 | 0.210 | | 0.25 | 0.168 | 0.266 | 0.234 | | 0.50 | 0.243 | 0.264 | 0.284 | | 0.75 | 0.285 | 0.286 | 0.325 | | 1.00 | 0.329 | 0.305 | 0.350 | | 1.25 | 0.351 | 0.329 | 0.370 | | 1.50 | 0.361 | 0.363 | 0.388 | | 1.75 | 0.360 | 0.387 | 0.392 | | 2.00 | 0.367 | 0.412 | 0.400 | | 3.00 | 0.368 | 0.507 | 0.398 | | 4.00 | 0.368 | 0.568 | 0.398 | Table 13: The mean loss report delay, measured at the sender. As can be seen from Table 13 the delay increases in general with an increasing value of l. However a similar effect as for the feedback suppression performance is visible: from a certain threshold, the additional increase in delay is not significant anymore. The threshold is environment dependent and seems to be related to the threshold, where the feedback suppression gain would not increase anymore. 8.3 Summary of "l" investigations We have shown that theoretically the performance of the feedback suppression mechanisms is increasing with an increasing value of l. The same applies for the report delay, which increases also with an increasing l. This leads to a threshold where both the performance and the delay does not increase any further. The threshold is environment dependent. So finding an optimum value of l is not possible because it is always a tradeoff between delay and feedback suppression performance. With l=0.5 we think that a tradeoff was found that is acceptable for typical applications and environments. Burmeister et al. Expires May 2002 22 Extended RTP Profile for RTCP-based Feedback November 2001 - Results of the Timing Rule Simulations - 9 Applications Using AVPF NEWPRED is one of the error resilience tools, which is defined in both ISO/IEC MPEG-4 visual part and ITU-T H.263. NEWPRED achieves fast error recovery using feedback messages. We simulated the behavior of NEWPRED in the network simulator environment as described above and measured the waiting time statistics, in order verify that the extended RTP profile for RTCP-based feedback (AVPF)[1] is appropriate for the NEWPRED feedback messages. Simulation results, which present in the following sections, show that the waiting time is enough small to get the satisfactory performance of NEWPRED. 9.1 NEWPRED Implementation in NS2 The agent that performs the NEWPRED functionality, called NEWPRED agent, is different from the RTP agent we described above. Some of the added features and functionalities are described in the following points: Application Feedback The "Application Layer Feedback Messages" format is used to transmit the NEWPRED feedback messages. Thereby the NEWPRED functionality is added to the RTP agent. The NEWPRED agent creates one NACK message for each lost segment of a video frame, and then assembles plural number of NACK messages corresponding to the segments in the same video frame, into one Application Layer Feedback Message. Although there are two modes, namely NACK mode and ACK mode in NEWPRED [6][7], only NACK mode is used in these simulations. The parameters of NEWPRED agent are as follows: f: Frame Rate(frames/sec) seg: Number of segments in one video frame bw: RTP session bandwidth(kbps) Generation of NEWPRED's NACK Messages The NEWPRED agent generates NACK messages when segments are lost. a. The NEWPRED agent generates plural number of NACK messages per one video frame when plural number of segments are lost. These are assembled into one FCI message per video frame. If there is no lost segment, no message is generated and sent. b. The length of one NACK message is 4 bytes. Let num be the number of NACK messages in one video frame(1 <= num <= seg). Thus, 12+4*num bytes is the size of the low delay RTCP feedback message. Measurements We defined two values to be measured: Burmeister et al. Expires May 2002 23 Extended RTP Profile for RTCP-based Feedback November 2001 - Results of the Timing Rule Simulations - - Recovery time The recovery time is measured as the time between the detection of a lost segment and reception of a recovered segment. We measured this "recovery time" for each lost segment. - Waiting time The waiting time is the additional delay due to the feedback limitation of RTP. Fig.1 depicts the behavior of a NEWPRED agent when a loss occurs. The recovery time is approximated as follows: (Recovery time) = (Waiting time) + (Transmission time for feedback message) + (Transmission time for media data) Therefore, the waiting time is derived as follows: (Waiting time) = (Recovery time) - (Round-trip delay), where (Round-trip delay ) = (Transmission time for feedback message) + (Transmission time for media data) Picture Reference |: Picture Segment ____________________ %: Lost Segment /_ _ _ _ \ v/ \ / \ / \ / \ \ v \v \v \v \ \ Sender ---|----|----|----|----|----|---|-------------> \ \ ^ \ \ \ / \ \ \ / \ \ v / \ \ x / \ \ Lost / \ \ x / \ _____ v x / NACK v Receiver ---------------|----%===-%----%----%----|-----> |-a-| | |------- b -------| a: Waiting time b: Recover time (%: Video segments are lost) Fig.1: Relation between the measured values at the NEWPRED agent Burmeister et al. Expires May 2002 24 Extended RTP Profile for RTCP-based Feedback November 2001 - Results of the Timing Rule Simulations - 9.2 Simulation We conducted two simulations (Simulation A and Simulation B). In Simulation A, the packets are dropped with a fixed packet loss rate on a link between two NEWPRED agents. In Simulation B, packet loss occurs due to congestion from other traffic sources, i.e. ftp sessions. 9.2.1. Simulation A - Constant Packet Loss Rate The network topology, used for this simulation is shown in Fig.2. Link 1 Link 2 Link 3 +--------+ +------+ +------+ +--------+ | Sender |------|Router|-------|Router|------|Receiver| +--------+ +------+ +------+ +--------+ 10(msec) x(msec) 10(msec) Fig2. Network topology that is used for Simulation A Link1 and link3 are error free, and each link delay is 10 msec. Packets may get dropped on link2. The packet loss rates (Plr) and link delay (D) are as follows: D [ms] = {10, 50, 100, 200, 500} Plr = {0.005, 0.01, 0.02, 0.03, 0.05, 0.1, 0.2} Session band width, frame rate and the number of segments are shown in Table 14 +------------+----------+-------------+-----+ |Parameter ID| bw(kbps) |f (frame/sec)| seg | +------------+----------+-------------+-----+ | 32k-4-3 | 32 | 4 | 3 | | 32k-5-3 | 32 | 5 | 3 | | 64k-5-3 | 64 | 5 | 3 | | 64k-10-3 | 64 | 10 | 3 | | 128k-10-6 | 128 | 10 | 6 | | 128k-15-6 | 128 | 15 | 6 | | 384k-15-6 | 384 | 15 | 6 | | 384k-30-6 | 384 | 30 | 6 | | 512k-30-6 | 512 | 30 | 6 | | 1000k-30-9 | 1000 | 30 | 9 | | 2000k-30-9 | 2000 | 30 | 9 | +------------+----------+-------------+-----+ Burmeister et al. Expires May 2002 25 Extended RTP Profile for RTCP-based Feedback November 2001 - Results of the Timing Rule Simulations - Table 14: Parameter sets of the NEWPRED agents Figure3 shows the packet loss rate vs. mean of waiting time. A plotted line represents a parameter ID ( "[session bandwidth] - [frame rate] - [the number of segments] - [link2 delay]" ). E.g. 384k-15-9-100 means the session of 384kbps session bandwidth, 15 frames per second, 9 segments per frame and 100msec link delay. When the packet loss rate is 5% and the session bandwidth is 32kbps, the waiting time is around 400msec, which is just allowable for reasonable NEWPRED performance. When the packet loss rate is less than 1%, the waiting time is less than 200msec. In such a case, the NEWPRED allows as much as 200msec additional link delay. When the packet loss rate is less than 5% and the session bandwidth is 64kbps, the waiting time is also less than 200msec. In 128kbps cases, the result shows that when the packet loss rate is 20%, the waiting time is around 200msec. In cases with more than 512kbps session bandwidth, there is no significant delay. This means that the waiting time due to the feedback limitation of RTCP is neglectable for the NEWPRED performance. +------------------------------------------------------------+ | | Packet Loss Rate = | | Bandwidth | 0.005| 0.01 | 0.02 | 0.03 | 0.05 |0.10 |0.20 | |-----------+------+------+------+------+------+------+------| | 32k |130- |200- |230- |280- |350- |470- |560- | | | 180| 250| 320| 390| 430| 610| 780| | 64k | 80- |100- |120- |150- |180- |210- |290- | | | 130| 150| 180| 190| 210| 300| 400| | 128k | 60- | 70- | 90- |110- |130- |170- |190- | | | 70| 80| 100| 120| 140| 190| 240| | 384k | 30- | 30- | 30- | 40- | 50- | 50- | 50- | | | 50| 50| 50| 50| 60| 70| 90| | 512k | < 50 | < 50 | < 50 | < 50 | < 50 | < 50 | < 60 | | | | | | | | | | | 1000k | < 50 | < 50 | < 50 | < 50 | < 50 | < 50 | < 55 | | | | | | | | | | | 2000k | < 30 | < 30 | < 30 | < 30 | < 30 | < 35 | < 35 | +------------------+------+------+------+------+------+------+ Fig. 3 The result of simulation A 9.2.2. Simulation B - Packet Loss due to Congestion Burmeister et al. Expires May 2002 26 Extended RTP Profile for RTCP-based Feedback November 2001 - Results of the Timing Rule Simulations - The configuration of link1, link2, and link3 are the same as in simulation A except that link2 is also error-free, regarding bit errors. However in addition, some FTP agents are deployed to overload link2. See Figure 4 for the simulation topology. Link1 Link2 Link3 +--------+ +------+ +------+ +--------+ | Sender |------|Router|-------|Router|------|Receiver| +--------+ /|+------+ +------+|\ +--------+ +---+/ | | \+---+ +-|FTP|+---+ +---+|FTP|-+ | +---+|FTP| ... |FTP|+---+ | ... +---+ +---+ +---+ +---+ FTP Agents FTP Agents Fig4. Network Topology of Simulation B The parameters are defined as for Simulation A with the following values assigned: D[ms] ={10, 50, 100, 200, 500} 32 FTP agents are deployed at each edge, and totally 64 FTP agents are active. The sets of session bandwidth, frame rate, the number of segments are the same as in Simulation A (Table 14) We provide the results for the cases of 64 FTP agents, because these are the cases where packet losses could be detected stable. The results are similar to the Simulation A except for a constant additional offset of 50..100ms. This is due to the delay incurred by the routers buffers. 9.3 Summary of Application Simulations We have shown that the limitations of RTP AVPF profile do not generate such high delay to the feedback messages that the performance of NEWPRED is degraded in the sessions from 32kbps to 2Mbps. We could see that the waiting time increases with a decreasing session bandwidth and/or an increasing packet loss rate. Thereby it is not significant what the packet loss caused. Burmeister et al. Expires May 2002 27 Extended RTP Profile for RTCP-based Feedback November 2001 - Results of the Timing Rule Simulations - Congestion or constant packet loss rates behave similar. Still we see that for reasonable conditions and parameters the AVPF is well suited to support the feedback needed for NEWPRED. 10 Summary The new RTP profile AVPF was investigated regarding performance and potential dangers to the network stability. Simulations were conducted using the network simulator, simulating unicast and different sized multicast topologies. The results were shown in this document. Regarding the network stability, it was important to show that the new profile does not lead to any feedback explosion, or use more bandwidth as it is allowed. Thus we measured the bandwidth that was used for RTCP in relation to the RTP session bandwidth. We have shown that more or less exactly 5% of the session bandwidth is used for RTCP, in all considered scenarios. The scenarios included unicast with and without bit errors, different sized multicast groups, with and without errors or congestion on the links. Thus we can say that the new profile behaves network friendly in that sense that it uses only the allowed bandwidth that was assigned by RTP. Second we have shown that receivers using the new profile experience a performance gain. We have shown that especially RTP receiver that do have an RTT estimation to the sender gain from using the new profile. But also the other receivers could increase their performance. This was measured by the delay that the sender sees for the received feedback. Using the new profile this delay can be decreased by orders of magnitude. Third we investigated certain parameters of the new algorithms. We have shown that there does not exist an optimum value for those. The influence of the parameters is highly environment specific and a tradeoff between performance of the feedback suppression algorithm and the experienced delay has to be found. The values that are given in the draft seem to be reasonable for most applications and environments. Burmeister et al. Expires May 2002 28 Extended RTP Profile for RTCP-based Feedback November 2001 - Results of the Timing Rule Simulations - References [1] J.Ott, S.Wenger, S.Fukunaga, N.Sato, K.Yano, A.Miyazaki, K.Hata, R.Hakenberg, C.Burmeister: Extended RTP Profile for RTCP-based Feedback, Internet Draft, draft-ietf-avt-rtcp-feedback-00.txt, Work in Progress, July 2001. [2] H.Schulzrinne, S.Casner, R.Frederick, and V.Jacobson: RTP - A Transport Protocol for Real-time Applications, Internet Draft, draft-ietf-avt-rtp-new-10.txt, Work in Progress, July 2001. [3] H.Schulzrinne, S.Casner: RTP Profile for Audio and Video Conferences with Minimal Control, Internet Draft, draft-ietf-avt-profile-new-11.txt, Work in Progress, July 2001. [4] Network Simulator Version 2 - ns-2, available from http://www.isi.edu/nsnam/ns [5] C.Burmeister, T.Klinner: Low Delay Feedback RTCP - Timing Rules Simulation Results. Technical Report of the Panasonic European Laboratories, September 2001, available from http://www.pel.panasonic.de/ietf/docs/SimulationResults-A.pdf [6] ISO/IEC 14496-2:1999/Amd.1:2000, "Information technology - Coding of audio-visual objects - Part2: Visual", July 2000. [7] ITU-T Recommendation, H.263. Video encoding for low bitrate communication. 1998. [8] S. Fukunaga, T. Nakai, and H. Inoue, "Error Resilient Video Coding by Dynamic Replacing of Reference Pictures," IEEE Global Telecommunications Conference (GLOBECOM), pp.1503-1508, 1996. [9] Hideaki Kimata, Yasuhiro Tomita, Hiroyuki Yamaguchi, Susumu Ichinose, and Tadashi Ichikawa, "Receiver-Oriented Real-Time Error Resilient Video Communication System: Adaptive Recovery from Error Propagation in Accordance with Memory Size at Receiver," Electronics and Communications in Japan, Part 1, vol.84, no.2, pp.8-17, 2001. Burmeister et al. Expires May 2002 29 Extended RTP Profile for RTCP-based Feedback November 2001 - Results of the Timing Rule Simulations - Authors Addresses Carsten Burmeister Panasonic European Laboratories GmbH Monzastr. 4c, 63225 Langen, Germany mailto:burmeister@panasonic.de Rolf Hakenberg Panasonic European Laboratories GmbH Monzastr. 4c, 63225 Langen, Germany mailto:hakenberg@panasonic.de Akihiro Miyazaki Matsushita Electric Industrial Co., Ltd 1006, Kadoma, Kadoma City, Osaka, Japan mailto :akihiro@isl.mei.co.jp Jśrg Ott Universit„t Bremen TZI MZH 5180, Bibliothekstr. 1, 28359 Bremen, Germany {sip,mailto}:jo@tzi.uni-bremen.de Noriyuki Sato Oki Electric Industry Co., Ltd. 1-2-27 Shiromi, Chuo-ku, Osaka 540-6025 Japan mailto:sato652@oki.co.jp Shigeru Fukunaga Oki Electric Industry Co., Ltd. 1-2-27 Shiromi, Chuo-ku, Osaka 540-6025 Japan mailto:fukunaga444@oki.co.jp Burmeister et al. Expires May 2002 30