INDUSTRIAL ENGINEERING AND
OPERATIONS RESEARCH
PRESENTS
IEOR MONDAY SEMINAR
APRIL 24, 2006
RESOURCE ALLOCATION AND OPTIMIZATION IN SERVICE NETWORKS
Kevin Ross
Graduate Program Director
Technology and Information Management
UC
ABSTRACT
Several modern systems in call centers, communication networks and flexible
manufacturing require the dynamic scheduling of resources in changing
environments. These are characterized by multiple types of requests arriving
simultaneously to the system at widely different and often unknown rates. A
network manager, or distributed group of them, allocates a finite pool of
resources to trade off alternatives such as serving a few jobs at a high rate
or serving many jobs concurrently at lower rates.
We present a framework for modeling and addressing some core
issues including scheduling, stability, scalability and load balancing. The
focus will be on projective cone scheduling (PCS) algorithms, which allocate
service to several queues based on observed backlogs. They maximize throughput
and can shape load balancing performance according to priorities. The results
apply to an extremely general model of service systems including networks where
the service and arrivals to each queue are interdependent and jobs can be
forwarded through a sequence of servers. Several interesting variations will be
introduced including networks with finite buffers and differential pricing of
service levels.
TIME AND LOCATION:
REFRESHMENTS
WILL BE SERVED @