Open MPI logo

MTT Devel Mailing List Archives

  |   Home   |   Support   |   FAQ   |   all MTT Users mailing list

From: Jeff Squyres \(jsquyres\) (jsquyres_at_[hidden])
Date: 2006-06-30 18:46:10


Andrew and I got on the phone to talk about the database stuff that
we've been discussing on the mailing list. There were several
miscommunications, but I think we understand each other now. Here's the
important bits:

- mtt does currently submit one test result at a time (e.g., the result
of a single intel test). It certainly can be adjusted to submit them
all at once, but right now it submits them individually.
- a perfbase "run" is actually a unique *table* in the database that
holds the variant data (not a *row* in an existing table, as I thought).

Hence, submitting 400+ "runs" for a single invocation of the intel test
suite would be so inefficient that we need to consider something better
(especially since we'll be invoking the intel test suite potentially
many times a night at each testing site -- so it's not just that we'll
be creating boatloads of tables; we'll be creating N time boatloads of
tables).

So for the moment, we do need to submit all the intel tests as a single
"run", meaning that we have to submit them to the database all at once
(how we upload them to the server is a different issue). This
unfortunately means no partial results.

For the immediate need, I think that this is ok because we're focusing
on nightly tests where partial results aren't as important. But this is
definitely something that we should ponder further as MTT becomes more
pervasive (e.g., during a release cycle when we might be running MTT
multiple times a day). It is probably even worth talking to the
perfbase author about this issue [...starting to compose a mail to the
perfbase mailing list...].

-- 
Jeff Squyres
Server Virtualization Business Unit
Cisco Systems