Network bandwidth requirements for scalable on-demand streaming

Authors:

• Yanping Zhao Asempra Technologies, San Jose, CA
• Derek L. Eager Department of Computer Science, University of Saskatchewan, Saskatoon, Canada
• Mary K. Vernon Computer Sciences Department, University of Wisconsin-Madison, Madison, WI

Complete Citation

Zhao, Y., Eager, D. L., and Vernon, M. K. 2007. Network bandwidth requirements for scalable on-demand streaming. IEEE/ACM Trans. Netw. 15, 4 (Aug. 2007), 878-891. DOI= http://dx.doi.org/10.1109/TNET.2007.893886

Abstract

Previously proposed streaming protocols using broadcast or multicast are able to deliver multimedia files on-demand with required server bandwidth that grows much slower than linearly with request rate, or with the inverse of client start-up delay. The same efficiencies can be achieved for network bandwidth if delivery is over a true broadcast channel. This paper considers the required network bandwidth for on-demand streaming over multicast delivery trees. We consider both simple canonical delivery trees, and more complex cases in which delivery trees are constructed using both existing and new algorithms for randomly generated network topologies and client site locations. Results in this paper quantify the potential savings from use of multicast trees that are configured to minimize network bandwidth rather than the latency to the content server. Further, we determine the network bandwidth usage of particular immediate service and periodic broadcast on-demand streaming protocols. The periodic broadcast protocol is able to simultaneously achieve close to the minimum possible network and server bandwidth usage.

Annotations

This paper investigates the bandwidth requirements for streaming media, assuming some form of multicast is enabled. The authors investigate the performance of various tree-building algorithms under randomly generated topologies using INet2.1 (?). Basically the authors demonstrate that many of the protocols can achieve a O(log K) growth in bandwidth requirements for smaller networks. For larger networks generated using the INet2.1 you do not end up with an even distribution of client nodes. Therefore to ensure scalability of the content it is better to clients to not use the shortest path tree join, but a more efficient join that may introduce latency but will allow for scalabilty.

This work is well done in that there are heavy mathematical derivations deriving the Big-O bounds for the efficiency of the different tree structures, but I guess I think that the results that they presented are nothing surprising. I was just surprised to see this in TON 2007.

-- DavidSalyers - 31 Oct 2007