90 likes | 108 Views
This paper presents SplitStream, an innovative approach to multicast streaming that splits video streams into smaller segments and uses separate multicast trees for each segment. The method aims to improve fault tolerance and avoid overburdening specific nodes. However, implementing this approach requires complex tree-building algorithms and poses challenges in ensuring each peer receives all stream segments.
E N D
SplitStream: high-bandwidth multicast in cooperative environments M. Castro, P. Druschel, A. Kermarrec,A. Nandi, A. Rowstron and A. Singh
Paper overview • Proposes a better alternative for conventional tree multicast • Split video stream into k streams • Multicast each stripe using a separate tree • Ensure that each peer node is at interior node in only one tree and a leaf in all other trees.
Single tree-based multicast Three nodes out of seven do all the work!
Bad choice for P2P multicasting Interior nodes do all the work Leaf nodes do nothing Does not tolerate interior node failures Even worse for high bandwidth applications Most peers do not have enough resource to transmit the data to act as leaf nodes Single tree-based multicast
Splitting streams (I) • Divide single video stream into k smaller split streams • Can reconstitute the video stream by merging the split streams
Splitting streams • Use a different multicast tree for each split stream
Fault-tolerance • New scheme is less fault-tolerant because data come from different sources • To make it fault tolerant, use • Multiple Description Coding (MDC) • Loss of one or more split streams will result in a loss of video quality • Erasure coding • Introduce one or more redundant split streams (same idea as RAID level 5 or 6)
The main issue • Find a way to build a forest containing k trees such that • Each tree distribute a distinct split stream • Each peer receives all k split streams • Must belong to each of the k trees • Each peer is an internal node in at most one tree • Most difficult condition to enforce • Very complex algorithm
Conclusions • It works • Requires a very complex tree building algorithm