Newsgroups: comp.parallel,comp.parallel.pvm,comp.parallel.mpi
From: Tae Ho Cho <cho@ece.arizona.edu>
Subject: Scheduler for Heterogeneous DCE
Organization: U of Arizona Electrical and Computer Engineering
Date: Mon, 9 Jan 1995 15:17:44 GMT
Message-ID: <3eo470$lnb@news.CCIT.Arizona.EDU>

Hi there,

We are currently developing a dynamic scheduler for
Heterogeneous Distributed/Parallel computing environment.
For our research purpose, we would like to model the
performance of a workstation. For instance, for a
given CPU/memory specifications, we want to measure
how long it takes to execute a certain task
(task size varys).  We might create more than one process in
one workstation.

So far we have figured out that it's not feasible to trace
the computing behavior of multiple processes in workstation
(to measure execution time (wall clock)) and to create
precise analytic model.

However, we can measure the execution time of a job(or jobs)
in certain workstation (I/O behavior only).  With this data,
we might be able to model the workstation (using Neural Net).

Does somebody have done some research on this problem?

Any comments/information/papers are pretty much appreciated.

regards,


Tae H. Cho
AI and Simulation Group
Dept. of Electrical and Computer Engineering
University of Arizona
Tucson, AZ 85721




