Title:

Investigating Inaccuracy in User Runtime Estimates and "The Padding
Hypothesis"

Speaker: 

Cynthia Bailey Lee

Abstract:

Computer system batch schedulers typically require information from the
user upon job submission, including a runtime estimate. Inaccuracy of
these runtime estimates, relative to the actual runtime of the job, has
been well documented and is a perennial problem mentioned in the job
scheduling literature. Typically users provide these estimates under
circumstances where their job will be killed after the provided amount of
time elapses. As a result, many believe the inaccuracy seen in requested
times is due to "padding" users add to avoid any possibility of having
their jobs killed. This assumes that users have an accurate (pre-padded)
estimate in mind, which they could provide, absent a job-kill threat. We
test whether this "Padding Hypothesis"  holds for actual users of a San
Diego Supercomputer Center system. We also examine system design
implications of estimate uncertainty and related user behaviors