Voice traffic is smooth, so it makes sense to talk about the utilization rate or the spare capacity of a line carrying voice traffic. But data traffic is self-similar, so it's not even meaningful to talk about average utilization or peak rates or anything really.

	 :	Even when individual computers limit their data transfer speeds, the resulting traffic is not as nicely behaved as voice traffic. It is now widely accepted that data traffic is self-similar [LelandTWW] (see [FeldmannGWK] for latest results and more complete references). This means that as transfers from many sources are aggregated, there is some smoothing, but much less than on the voice network. It seems that there are fundamental limitations on the efficiency that can be achieved on data networks.

	 :	The work on self-similarity of data traffic shows that the usual procedure of looking at just 5-minute or 1-hour averages of traffic is not adequate to understand what goes on. One should study traffic on millisecond time scales, but that is currently done only in a few experimental setups. Networks are engineered based on cruder averages, and the usual rules one hears about in high quality networks are of the form "a T1 link has to be upgraded if hourly averages exceed 50% of the capacity over more than 5% of the business hours." For Internet backbones, a common rule [Gareiss] is that "during peak periods, an ISP should have at least 30 percent to 40 percent of spare bandwidth. The good news is that most providers have 50 percent or more." (Unfortunately, there are many subtleties in defining spare bandwidth, so it is hard to interpret these claims precisely).

	 :	- AndrewOdlyzko http://www.dtc.umn.edu/~odlyzko/doc/network.utilization.pdf

	 :	[LelandTWW] W. E. Leland, M. S. Taqqu, W. Willinger, and D. V. Wilson, "On the self-similar nature of Ethernet traffic (extended version)", ''IEEE/ACM Trans. Networking'' 2 (1994), 1-15. Available at http://www-net.cs.umass.edu/cs691s/leland.pdf

	 :	[FeldmannGWK] Anja Feldmann, Anna C. Gilbert, W. Willinger, and T. G. Kurtz, "The changing nature of network traffic: Scaling phenomena", ''Computer Communication Review'', 28, no. 2 (April 1998), pp. 5-29. Available at http://www.math.wisc.edu/~kurtz/papers/ccr98_scaling_long.ps.

	 :	[Gareiss] Robin Gareiss, Is the Internet in trouble?, ''Data Communications'', Sept. 21, 1997, pp. 36-50. Available at http://web.archive.org/web/19990427211433/http://www.data.com/roundups/trouble.html.

More on the topic:

* "The self-similarity and long-range dependence in networks" web site at http://www.cs.bu.edu/pub/barford/ss_lrd.html
* Anja Feldman's homepage: http://www.net.informatik.tu-muenchen.de/~anja/
* Anna Gilbert's homepage: http://www.math.lsa.umich.edu/~annacg/

----
CategoryInternet