EECS 290Q: Project Abstracts

Analysis of the Generalized Processor Sharing Scheduling Discipline using Statistical Bounds
Tuan Le

A. Parekh and R. Gallager analyzed the Generalized Processor Sharing (GPS) scheduling discipline and established worst-case deterministic bounds for networks of servers. However, in applications where some loss is acceptable, deterministic bounds can lead to low network utilization if they are used to regulate access to the network. Hence, soft statistical bounds are beneficial.

By appropriately bounding the probability distribution functions of the source traffic streams, we can acquire statistical bounds for parameters such as queue lengths and delays. I will present a method of bounding the source traffic and show how results for a single GPS server and multiple GPS servers can be obtained. Since the source traffic characterization is fairly general, the same techniques can be applied to other scheduling disciplines.

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