Exploration of RTP Circuit Breaker with Applications to Video Streaming Nazila Fough, Fabio Verdicchio, Gorry Fairhurst Electronic Research Group, University of Aberdeen, Aberdeen, UK {nazila.fough, fverdicc, g.fairhurst}@abdn.ac.uk Abstract—Live multimedia streaming is becoming one of the dominant sources of internet traffic, much of which is sent over best-effort networks, i.e. along paths with a wide variety of characteristics. The multimedia traffic should be transmitted using a robust and effective congestion control mechanism to protect the network from congestion collapse. The RTP Circuit Breaker (RTP-CB) is a candidate solution that causes a sender to cease transmission when RTCP message feedback indicates excessive congestion. This paper studies RTP/UDP video traffic and the impact of its bursty behavior on the network. It considers the potential limitations of using a RTP-CB with video traffic. We found that the bursty nature of a typical video flow can cause the RTP-CB to either prematurely cease transmission or to react too late. To reduce the likelihood of this happening, we suggest the use of a smoothing buffer in conjunction with the RTP-CB and propose design criteria for this buffer. Our experiments confirm the effectiveness of the proposed approach for different video streams. Index Terms—RTP/UDP video traffic, RTP-Circuit Breaker I. INTRODUCTION Multimedia streaming, video conferencing and other real- time multimedia applications are becoming increasingly popular and are expected to dominate Internet traffic in the near future. Deployment of applications using the RTCWEB protocol [1], and expected increase in media streaming pose considerable challenges for sharing the available capacity with other traffic in existing wired/ wireless network infrastructure. To prevent network congestion, there is an urgent need for these methods to support some form of congestion control. However, at the time of writing this paper, there is no generally accepted congestion-control method for these media flows [2]. A TCP flow adapts its rate based on a congestion control algorithm utilising feedback received over the network. Hence a controlled flow adapts its sending rate, by taking into consideration the state of the network path, and reacts to avoid impending congestion. Internet video traffic is usually carried using RTP/UDP, which does not implement congestion control at the transport layer, with the rate simply determined by the video codec settings. Hence uncontrolled UDP video traffic has the potential to induce significant congestion. Furthermore, when controlled (e.g. TCP [3]) and uncontrolled (e.g. UDP [4]) multimedia flows share a link, the well behaved flows reduce their rate and capacity usage, while the unrestricted ones unfairly dominate the bottleneck usage and retain the potential to congest the bottleneck, thus damaging all the flows. The RTP Circuit Breaker (RTP-CB), recently proposed in an Internet Engineering Task Force (IETF) internet-draft [5], is a technology that is expected to protect the network from excessive congestion. A media sender that receives notification of congestion should invoke the RTP-CB to cease transmission to avoid further congestion. It is expected that well-controlled RTP applications using best-effort networks will be able to operate without triggering the RTP-CB, while misbehaving flows will be blocked to prevent congestion and avoid damaging other users. In this paper we review the relevant features of a video streaming session (Section II). We link these features to the effective use of the RTP-CB as a safety switch for a video flow (Section III). We highlight potential problems with the RTP- CB and propose a solution based on a pacing buffer (Section IV). To our knowledge this study has not been performed before. We provide preliminary experimental evidence that support the effectiveness of the proposed scheme (Section V) and discuss our findings (Section VI). II. VIDEO STREAMS OVER NETWORKS We begin by reporting the transmission profile of a typical video stream. The sample video sequence is denoted as Test1, it is one of the sequences considered in this paper and described in Table 1 (Section V). The streaming system architecture comprises a streaming server (acting as media provider) a network emulator (simulating a bottleneck link with limited capacity and a drop-tail buffer) and a video client (representing the user). The testbed is further described in Section V. The transmission profile for a typical eight-second segment of the sequence is shown in Figure 1. Each point in the figure represents the average transmission rate over a one second interval. The sequence is encoded at a nominal rate of 1 Mb/s and the points in the graph lay around this value. For the same video segment, Figure 2 shows the instantaneous transmission profile: each point in the figure represents the average transmission rate over a short interval of 5 ms. The figure shows the presence of distinctive spikes reaching instantaneous rate more than ten times above the nominal rate. We report that these spikes are a characteristic of every video streaming session we considered. Streaming over a limited capacity link following the transmission profile of Figure 2 can severely impact the quality of the received video. For example, we streamed the video sample Test1 over a path with a capacity of 1.5 Mb/s and a bottleneck buffer size of approximately 10 packets. The link capacity is greater than the nominal rate of the video (1 Mb/s) hence the moderate size of the buffer, in line with current trends [6], would seem appropriate. We would therefore expect The work of N. Fough is supported by EPSRC. ISBN: 978-1-902560-27-4 © 2013 PGNet