[comp.parallel] Parallel Execution Tracefiles?

gannon@iuvax.cs.indiana.edu (Dennis Gannon) (09/26/90)

A Call for participation:

a small group of people has decided to organize an
"birds of a feather" session at Supercomputing '90 in New York
this year on the topic described below.


**********************************************************

SUBJECT:  Standardizing parallel trace formats
TIME:	  7:00 - 10:00 PM
DATE:	  Thursday, November 15
ORGANIZERS:  Cherri M. Pancake (Auburn University)
	     Dennis Gannon (Indiana University)
	     Donna Bergmark (Cornell National Supercomputer Facility)
	     Sue Utter (Cornell National Supercomputer Facility)

ABSTRACT:
Although facilities to trace the execution of parallel programs
are widely available, there is no standardized format for trace
output.  The types of events monitored and the information reported
for each event vary considerably from one vendor to the next.
This situation complicates the development of tools for trace
analysis or replay.  To port a tool across parallel systems,
the developer must implement a series of translators to handle
the diverse formats, perhaps modifying the tool to reflect
the presence or absence of certain information.

A standardized trace format, supported across vendor lines,
would make it possible to adopt tools to a wider range of
systems at lower cost.	It would also encourage a broader
distribution of trace-based tools.  This session will address
the requirements to be met by such a standard.	Recommendations
on the form and content of trace records will be presented
for discussion.
*************************************************************

This is, of course, not a new idea.  We understand several groups
have recently proposed such standards and we would like to include
them in our discussions. We request that anybody with knowledge of such proposals
please contact me (gannon@iuvax.cs.indiana.edu) AND Sue Utter
(psu@cornellf.tn.cornell.edu).	Also, to get the discussion started
for this sessions, I would like to see comments posted to this notes
file as to merit or folly of such a standardization plan.  This is
especially important if you are not going to the New York meeting
but you feel you feel you have a lot to say on this topic.

thanks
Dennis Gannon
CICA & DCS 
Indiana University