EECS 290Q: Project Abstracts

Switch Architectures and QoS
Cliff Cordeiro

In this class, when we've discussed QoS, it has mostly been in regards to delay and bandwidth. We've assumed ideal switching and output queueing. Real switches are not ideal, however. There are cell losses which must be kept to a minimum, internal conflicts which should be avoided, unfairness which should be compensated for, and sometimes no provision for any form of per VC QoS guarantees (imagine, ATM without QoS). It is imperative to create swtiches which are as close to the ideal as possible and which implement the largest subset of desirable features (including multicast). Depending on the architecture of the switch, however, some specifications are more difficult or even infeasible to implement. This talk will cover the main types of switch architectures, the QoS issues involved in designing a switch, and the QoS tradeoffs involved in switch design.

UCB Networking / Matt Siler siler@eecs.berkeley.edu/ April 19, 1996