Monday, April 6, 1998
2:00 PM (refreshments 1:45)
Grier Room, Room 34-101A
EECS Special Seminar
Abstract
Recently, there has been an enormous growth in the volume and diversity of traffic on the Internet. An important consequence of this growth is that early models of network traffic processes, as Poisson or Poisson-like processes, have been made obsolete. A sobering conclusion of many recent studies is that it is not yet possible to accurately model or simulate a trace of actual network traffic. Furthermore, new applications, protocols or data-coding mechanisms may bring new traffic types in the future. As a result, the field of networking faces two problems simultaneously: (1) There is a need for newer, more general queueing models in stochastic network theory; and (2) there is a need for a faster switching/routing infrastructure that can offer differentiated quality-of-service (QoS) to individual users and work well regardless of the nature of the offered traffic.
In this talk I will present recent work on both sets of problems. First, I will briefly describe a series of simulations that initiated the study of general (non-Poisson) queueing networks, and the resulting conjectures about the behavior of such networks. I will show how these conjectures are resolved, leading to a general theory of stochastic networks.
Second, I will describe the solution of the so-called "problem of speedup" in high-performance switch and router design. Specifically, I will exhibit scheduling algorithms that allow one to inexpensively build very high-speed switches capable of providing differentiated QoS to individual users. Our approach allows us to exactly mimic the desirable performance of output-buffered switches under arbitrary input traffic statistics, without having to incur the associated cost of a high memory bandwidth.
|
Modified: Mar 19, 1998|
Current events|
Your comments
and inquiries are welcome.