My area of research is Workload Generation for Parallel Computers. As of now, I am
devloping a program that generates synthetic workload for parallel computers using
the statistical models of the real workload traces, collected from various
supercomputing sites. Basically, my work involves statistical modeling of the
real workload traces using various statistical distributions like Gamma and Weibull
distributions and then generation of a synthetic workload which will represent the
real workload and at the same time, will be fleixble enough to allow changes
in the workload characteristics with changes in the modeling parameters.
The workload, the scheduling algorithm is subjected to during testing, plays an important
role in the design of job schedulers for parallel computers.
More realistic the workload, better will be the design of the algorithm and hence
the efficiency of the job scheduler.
A number of reseachers have been working in this area. My work is based on the
approach suggested by Dr. Feitelson (The Hebrew University, Israel). As of now
my program is able to model workload parameters like run-time, job size. I am
working on modeling of the jobs arrivals process which confirms to the daily
jobs-arrival cycle as well as the weekly cycle.