Citation Context

...etwork round-trips in a centralized system such as Napster or Google but O(log N) (or more) hops (where N is the number of nodes in the system) in a distributed system such as one based on DHTs [33], =-=[36]-=-. Some of these goals (appear to) conflict with each other, so we prioritize them as follows. Since resilience is our overall objective, we choose to focus on building short and diverse trees with sho...

Citation Context

...O(1) network round-trips in a centralized system such as Napster or Google but O(log N) (or more) hops (where N is the number of nodes in the system) in a distributed system such as one based on DHTs =-=[33]-=-, [36]. Some of these goals (appear to) conflict with each other, so we prioritize them as follows. Since resilience is our overall objective, we choose to focus on building short and diverse trees wi...

Citation Context

...ing and path diversity in turn. A. Application-level Multicast The deployment of IP multicast [16], [17], especially at the inter-domain level, has been slow due to technical and operational concerns =-=[14]-=-. This has spurred the development of application-level multicast schemes where end-hosts (clients and/or servers) perform the role of “routers”. Narada [14] and Scattercast [11] build application-lev...

Citation Context

... median outgoing bandwidth was 212 Kbps (these were measured using TCP transfers to/from well-connected servers). This asymmetry is consistent with anecdotal evidence and previous measurement studies =-=[34]-=-, and clearly indicates that the outgoing bandwidth of peers is likely to be the constraining factor in CoopNet. The median P2P latency (ping time) we measured was 80 ms. But the median latency even a...

Citation Context

...f tree construction within the constraints imposed by node bandwidth and the desire for short trees. The 4 This statement is not strictly true because Internet connectivity is not strictly transitive =-=[4]-=-. A node may lose direct connectivity to the root and hence be unable to exchange tree management messages with the root, yet it may be able to receive the data stream routed via its ancestors (i.e., ...

Citation Context

...to competition from other traffic. Thus we need a solution that accommodates bandwidth heterogeneity and congestion control. An elegant framework proposed in the literature is based on layered coding =-=[24]-=-. The idea is to encode the streaming content into layers so that a client can (dynamically) choose how many layers to subscribe to depending on the currently available bandwidth. Inspired by this, we...

Citation Context

...is disseminated to all members of the mesh). However, these could be used in the context of CoopNet for communication among a small, stable set of distributed servers. An alternative approach is NICE =-=[6]-=-, which uses a hierarchy to scale better than a mesh-based protocol. However, NICE is not optimized for a high rate of node churn. Joins require 14O(log(N)) network round-trips, where N is the size o...

Citation Context

...n only so long as the user is interested in the content. It stops forwarding traffic when the user tunes out. This requirement makes CoopNet fundamentally different from many other P2P systems (e.g., =-=[10]-=-) where nodes are expected to route traffic so long as they are online, even if they are themselves not interested in the corresponding content. We also insist that a node only contribute as much (ups...

Citation Context

...2.58X and 3.18X), respectively, off the optimal. While these results are encouraging, it would also be interesting to compare against more sophisticated network distance estimation algorithms such as =-=[26]-=-. 4) Parameter Settings: The parameters for our simulation experiments are set as listed in Table III. The stream bandwidth and the outgoing bandwidth available at each node are set to 160 Kbps each, ...

Citation Context

... tree due to node failures can take up to 30 seconds to heal. In contrast, CoopNet exploits the availability of a stable and resourceful server to optimize these operations. In ALMI [30] and Overcast =-=[19]-=-, a central node coordinates tree management, as in CoopNet. In ALMI, a centralized session controller gathers peer-to-peer ping data to perform a bounded-degree minimum spanning tree computation and ...

Citation Context

...our work spans multiple areas. We discuss work on application-level multicast and that on source coding and path diversity in turn. A. Application-level Multicast The deployment of IP multicast [16], =-=[17]-=-, especially at the inter-domain level, has been slow due to technical and operational concerns [14]. This has spurred the development of application-level multicast schemes where end-hosts (clients a...

Citation Context

...ource Coding and Path Diversity Several researchers have advocated the use of source coding, possibly in conjunction with path diversity, to make data transfer robust to packet loss. Digital Fountain =-=[8]-=- uses Tornado codes (a form of erasure coding) coupled with multiple multicast groups to distribute files scalably to a heterogeneous population of clients. The source transmits the coded blocks repea...

Citation Context

...coding schemes have been investigated over the years. For an overview see [18]. A particularly efficient and practical system is based on layered audio or video coding [29], [23], Reed-Solomon coding =-=[37]-=-, priority encoded transmission [3], and optimized bit allocation [15], [32], [25]. In such a system the audio and/or video signal is partitioned into groups of frames (GOFs), each group having a dura...

Citation Context

...rt trees to minimize the likelihood of disruption. Recent work has leveraged the scalable routing substrate provided by distributed hash tables (DHTs) to build efficient multicast trees (e.g., Bayeux =-=[39]-=-, Scribe [10]). It is unclear how well these perform in the face of a high rate of node churn, especially since the data structures needed for efficient routing are updated lazily. Furthermore, a fund...

Citation Context

... synonymously in this paper. 3 We term these as “failures” although the node may not have actually failed. 2in network paths. The streaming content is encoded using multiple description coding (MDC) =-=[18]-=- and the descriptions are distributed over different trees. As our experimental results show, this approach significantly improves the quality of the received stream in the face of a high level of nod...

Citation Context

...trol scheme appropriate for situations where the last-hop links of end-hosts are prone to congestion, say due to competition from other applications running on the hosts. In a previous workshop paper =-=[28]-=-, we sketched the basic idea of CoopNet (viz., combining multiple distribution trees with MDC) and presented some preliminary analysis. This paper is substantially different in many respects. The tree...

Citation Context

...disruptions in the tree due to node failures can take up to 30 seconds to heal. In contrast, CoopNet exploits the availability of a stable and resourceful server to optimize these operations. In ALMI =-=[30]-=- and Overcast [19], a central node coordinates tree management, as in CoopNet. In ALMI, a centralized session controller gathers peer-to-peer ping data to perform a bounded-degree minimum spanning tre...

Citation Context

...y of the mesh. An efficient application-level multicast tree is formed by running a reverse path forwarding algorithm on the mesh. The choice of link metrics depends on the application. For instance, =-=[13]-=- proposes a combination of bandwidth and latency metrics for a conferencing application. It is interesting to note that the the set of links spanned by the multiple trees in CoopNet can also be viewed...

Citation Context

...nt to our work spans multiple areas. We discuss work on application-level multicast and that on source coding and path diversity in turn. A. Application-level Multicast The deployment of IP multicast =-=[16]-=-, [17], especially at the inter-domain level, has been slow due to technical and operational concerns [14]. This has spurred the development of application-level multicast schemes where end-hosts (cli...

Citation Context

...d over the years. For an overview see [18]. A particularly efficient and practical system is based on layered audio or video coding [29], [23], Reed-Solomon coding [37], priority encoded transmission =-=[3]-=-, and optimized bit allocation [15], [32], [25]. In such a system the audio and/or video signal is partitioned into groups of frames (GOFs), each group having a duration of T = 1 second or so, for exa...

Citation Context

... repeatedly and clients tune in until they have received a sufficient number of blocks for decoding. Such repeated transmissions, however, are not feasible in our live streaming context. Byers et al. =-=[7]-=- use Digital Fountain erasure coding technique and parallel downloads to take advantage of lateral bandwidth between peers (like P2P file sharing systems like KaZaa [1] do). The use of multiple trees ...

Citation Context

...e Construction: While randomization would result in a degree of tree diversity, the question is whether we can do better. We leverage the insightful observation made in the recent work on SplitStream =-=[9]-=- that the outgoing bandwidth constraint of nodes can be honored by making each node an interior node in just one tree. (That said, there are some crucial differences between SplitStream approach and o...

Citation Context

... operational concerns [14]. This has spurred the development of application-level multicast schemes where end-hosts (clients and/or servers) perform the role of “routers”. Narada [14] and Scattercast =-=[11]-=- build application-level meshes formed by connections among a subset of node pairs. The links in the mesh are monitored periodically to improve the quality of the mesh. An efficient application-level ...

Citation Context

...uting in (telephone) networks dates back to the late 1970s [18]. The application of this approach in the context of the Internet has received increasing attention in recent years. Apostopolous et al. =-=[5]-=- advocate the use of MDC and path diversity for on-demand streaming from a content distribution network. The idea is for the client to request distinct descriptions from two or more server nodes (akin...

Citation Context

...particularly efficient and practical system is based on layered audio or video coding [29], [23], Reed-Solomon coding [37], priority encoded transmission [3], and optimized bit allocation [15], [32], =-=[25]-=-. In such a system the audio and/or video signal is partitioned into groups of frames (GOFs), each group having a duration of T = 1 second or so, for example. Each GOF is then independently encoded, e...

Citation Context

...architecture. Figure 3 shows the architecture we have implemented. The input stream is from a layered codec; in our implementation, we use the PFGS codec (also known as the 6SMART codec) reported in =-=[38]-=-. The sequence of operations is as follows: 1) Frames in a GOF are partitioned into a set of data units that carry rate-distortion information. The prioritizer prioritizes and sorts these data units a...

Citation Context

...8]. A particularly efficient and practical system is based on layered audio or video coding [29], [23], Reed-Solomon coding [37], priority encoded transmission [3], and optimized bit allocation [15], =-=[32]-=-, [25]. In such a system the audio and/or video signal is partitioned into groups of frames (GOFs), each group having a duration of T = 1 second or so, for example. Each GOF is then independently enco...

Citation Context

...C since no redundancy is incurred. Therefore, in CoopNet, to deal with bandwidth heterogeneity and congestion in addition to node failure and packet loss, we have developed a novel layered MDC scheme =-=[12]-=- in which the descriptions are partitioned into layers such that if there is a choice, descriptions in the least important layers are dropped first. For simplicity, here we concentrate on two layers. ...

Citation Context

... is often the bottleneck. In contrast, in a P2P setting like CoopNet, the constrained upstream bandwidth at peers and the transience of the peers makes path diversity and MDC advantageous. Lee et al. =-=[21]-=- present a framework where feedback from an AIMD congestion control protocol (in the form of transmission rate and packet loss profile) is used to optimize an MDC coder. This is related to the MDC ada...

Citation Context

...see [18]. A particularly efficient and practical system is based on layered audio or video coding [29], [23], Reed-Solomon coding [37], priority encoded transmission [3], and optimized bit allocation =-=[15]-=-, [32], [25]. In such a system the audio and/or video signal is partitioned into groups of frames (GOFs), each group having a duration of T = 1 second or so, for example. Each GOF is then independentl...

Citation Context

...or a playback of a recording. that is webcasting a high school football game.) Furthermore, there is some evidence that even large sites (e.g., CNN) are moving away from CDNs to in-house server farms =-=[22]-=-. An alternative to these infrastructure-based solutions is endhost-based or peer-to-peer content distribution. 2 A P2P approach is attractive in this setting because the bandwidth available to serve ...

Citation Context

...Many multiple description coding schemes have been investigated over the years. For an overview see [18]. A particularly efficient and practical system is based on layered audio or video coding [29], =-=[23]-=-, Reed-Solomon coding [37], priority encoded transmission [3], and optimized bit allocation [15], [32], [25]. In such a system the audio and/or video signal is partitioned into groups of frames (GOFs)...

Citation Context

...What we need is an efficient way to pick a proximate parent for a node without requiring extensive P2P network measurements. We use the simple delay-coordinates based “GeoPing” technique, proposed in =-=[27]-=- for a somewhat different application (viz., determining the geographic location of Internet hosts). Each node maintains its “delay coordinates” of ping times to a small set of landmark hosts (say 10 ...

Citation Context

...ding. Many multiple description coding schemes have been investigated over the years. For an overview see [18]. A particularly efficient and practical system is based on layered audio or video coding =-=[29]-=-, [23], Reed-Solomon coding [37], priority encoded transmission [3], and optimized bit allocation [15], [32], [25]. In such a system the audio and/or video signal is partitioned into groups of frames ...

Citation Context

...MDC coder. This is related to the MDC adaptation in CoopNet. However, CoopNet focuses on a multicast setting rather than unicast, with a fixed transmission rate (for each layer). In a different paper =-=[31]-=-, the same authors present some preliminary ideas on applying MDC in a multicast setting. Their proposal is to have application-level proxies that re-encode the stream placed 15at bottleneck links. H...

Citation Context

...he U.S. We conducted both P2P and client-server measurements from these vantage points. We present here a brief overview of the results relevant to CoopNet; for more details on PeerMetric, please see =-=[20]-=-. We found that the median incoming bandwidth was 900 Kbps and the median outgoing bandwidth was 212 Kbps (these were measured using TCP transfers to/from well-connected servers). This asymmetry is co...