Resilient Peer-to-Peer Streaming*Venkata N. Padmanabhan, Helen J. Wang, Philip A. ChouMicrosoft Research{padmanab, helenw, pachou}@microsoft.comAbstractWe consider the problem of distributing “live” streaming mediacontent to a potentially large and highly dynamic population of hosts.Peer-to-peer content distribution is attractive in this setting becausethe bandwidth available to serve content scales with demand. A ke ychallenge, however, is making content distribution robust to peertransience. Our approach to providing r o bustness is to introduceredundancy, both in network paths and in data. We use multiple,diverse distribution trees to provide redundancy in network paths andmultiple description coding (MDC) to provide redundancy in data.We present a simple tree management algorithm that provides thenecessary path diversity and describe an adaptation framework forMDC based on scalable receiver feedback. We evaluate these usingMDC applied to real video data coupled with real usage tracesfr om a major news site that experienced a large flash crowd for livestreaming content. Our results show very significant benefits in usingmultiple distribution trees and MDC, with a 22 dB improvement inPSNR in some cases.I. INTRODUCTIONWe consider the problem of distributing “live” streamingmedia content from a server to a potentially large and highlydynamic population of interested clients. We use the term“live” to refer to the simultaneous distribution of the samecontent to all clients; the content itself may either be truly liveor a playback of a recording. Due to the lack of widespreadsupport for IP multicast (especially at the inter-domain level),the server may resort to unicasting the stream to individualclients. However, this approach only scales up to a point. Asurge in the client population, say due to a flash crowd, couldeasily overwhelm the server’s bandwidth.A range of solutions have been proposed in the literatureand employed in practice. The content provider could purchaseadditional bandwidth and install a (possibly distributed) clusterof servers. Alternatively, the services of a content distributionnetwork (CDN) such as Akamai could be used to achievethe necessary scaling, thereby relieving the content providerfrom the task of scaling their server site. However, theseapproaches may not be cost effective, at least for small ormedium sized sites, because the normal traffic levels may notbe high enough to justify the cost of purchasing additionalbandwidth or subscribing to the services of a CDN. In fact, thevolume of traffic at a small site, even during a flash crowd, maybe too low to be of commercial interest to a CDN operator.(Consider, for instance, a flash crowd that overwhelms a serverthat is webcasting a high school football game.) Furthermore,*Please visit the CoopNet project page athttp://www.research.microsoft.com/projects/CoopNet/ for additionalinformation, including a pointer to a more detailed paper [28].there is some evidence that even large sites (e.g., CNN) aremoving away from CDNs to in-house server farms [23].An alternative to these infrastructure-based solutions isend-host-based or peer-to-peer content distribution.1AP2Papproach is attractive in this setting because the bandwidthavailable to serve content scales with demand (i.e., the numberof interested clients). This is the basis for the CoopNet systempresented in this paper. CoopNet makes selective use of P2Pnetworking, placing minimal demands on the peers. The goalis only to help a server tide over crises such as flash crowdsrather than replace the server with a pure P2P system.There are a few key issues that need to be addressedin CoopNet. First, users may be wary of dedicating theirbandwidth to the common good, especially when ISPs chargebased on (upstream) bandwidth usage. We address this issuein CoopNet by insisting that a node participate in and con-tribute bandwidth for content distribution only so long as theuser is interested in the content. It stops forwarding trafficwhen the user tunes out. This requirement makes CoopNetfundamentally different from many other P2P systems (e.g.,[12]) where nodes are expected to route traffic so long asthey are online, even if they are themselves not interested inthe corresponding content. We also insist that a node onlycontribute as much upstream bandwidth as it consumes inthe downstream direction2. This creates a natural incentivestructure where a node may tune in to higher bandwidth (andbetter quality) content if and only if it is also willing andable to forward traffic at the higher rate. We do not, however,consider the enforcement issue (e.g., blocking free-riders) inthis paper.A second key issue is that the nodes in CoopNet areinherently unreliable. The outgoing stream from a node maybe disrupted because the user tunes out, the node crashes orloses connectivity, or simply because the upstream bandwidthis temporarily used up by a higher-priority user task (e.g.,sending out an email with large attachments)3. The traditionalapproach to end-host-based application-level multicast, whichinvolves constructing a single distribution tree, is vulnerableto such failures because the descendants of the failed nodesmight experience severe disruption until the tree is repaired(or the failed nodes are revived). Parent-driven retransmission1We use the terms end-host-based multicast and peer-to-peer multicastsynonymously in this paper.2This restriction only applies to the total bandwidth in and out of a nodeaggregated over all trees. Thus the individual trees will still be “bushy”, asexplained in Section II-B3We term these as “failures” although the node may not have actually failed.Proceedings of the 11th IEEE International Conference on Network Protocols (ICNP’03) 1092-1648/03 $17.00 © 2003 IEEE(ARQ) is not a good fit because we are concerned with thefailure of the parent node itself, not just network packet drops.So we address the robustness issue in CoopNet by introducingredundancy, both in network paths and in data. Multiple, di-verse distribution trees spanning the set of participating nodesare constructed, thus providing redundancy in network paths.The streaming content is encoded using multiple descriptioncoding (MDC) [19] and the descriptions are distributed overdifferent trees. As our experimental results show, this approachsignificantly improves the quality of the received stream in theface of a high level of node churn.The use of
View Full Document
Unlocking...