'Geometry' of Resource Allocation and Load
Balancing in Computing/Communication Systems
Nick Bambos
Stanford University
Wednesday, November 07, 2007
4:30 - 5:30 PM
Terman Engineering Center, Room 453
Abstract:
We consider a processing system having several parallel buffers, where
classes of jobs/tasks are queued up, while awaiting execution. At each point
in time, the system can be set to one of several available service modes;
each queue then receives service at a mode-dependent rate. Modes correspond
to feasible configurations of various processing resources. The model unifies
resource allocation situations in autonomic computing, wireless transmitter
power control, packet switching, etc. Taking a 'geometric' perspective, we
discuss some new general families of algorithms, which maximize the
throughput by dynamically choosing service modes in response to observed
buffer backlogs. We also discuss issues pertaining to jitter and power
control captured by this model.