Speaker: Adam Wierman

Time: Wednesday 12-1pm

Place: NSH 1507

Title: Understanding the effects of SMART scheduling

Abstract:

Recently, there have been a number of "scheduling success stories" where
computer applications have provided improved performance by simply
adjusting the way requests are queued.  In applications such as web
servers and routers, the heuristic of biasing towards small jobs has led
to many of these successes.  However, the policies that are implemented by
practitioners do not match the idealized policies studied in theory.  In
my thesis, my goal is to illustrate that it is possible to bridge this gap
by analyzing classes of scheduling policies instead of individual
policies.  In this talk, I will present one example of such a class: the
SMART class, which includes a wide range of policies that prioritize jobs
with small sizes.  We will show that the SMART class is 2-competitive with
respect to mean response time.  Further, we will both derive stochastic
bounds on the behavior of SMART policies and derive the asymptotic tail
behavior of the SMART class.