Systems Seminar
Fast Jackson Networks
Prof. Yuri Suhov
Cambridge University,
Visiting University of California, Santa Barbara
Abstract
The famous Jackson model remains till now a principal example
of a queueing network admitting (albeit partially) an exact
solution: the stationary distribution for the queue-size
(the number of tasks in the nodes), if it exists,
is a product of marginal geometric distributions,
with exponentially decreasing tails. The talk focusses
on a rigorous analysis of a class of Jackson-type
queueing networks where a principle
of dynamic routing is introduced: a task joins
a queue that is shortest in a randomly selected
sample. Such a modification provides an essential
improvement in network's performance, both in terms
of the conditions of existence of a stationary
distribution and its tail asymptotics. For example,
the stationary queue-size distribution becomes a product
of marginal distributions with superexponentially
decreasing tails.
NOTE SPECIAL TIME AND PLACE:
Wed., Mar. 24, Noon-1:00 pm in 2421 Engr. Hall.
SYSTEMS SEMINAR WEB PAGE:
http://www.cae.wisc.edu/~gubner/seminar/