Randy Stiles (stiles++at++aic.lockheed.com)
Thu, 28 Mar 1996 11:21:02 -0800
We have experienced the same phenomenon here on our
Indigo2 Impacts, where Performer 2.0 in multi-process can die horribly
and kick us out to the login prompt, except it is when we have
other applications running as well as Performer, which
are also memory intensive.
I believe this has to do with swap space when running multi-process,
but I am not certain. When mult-processing, more memory
is used for the memory copies of Performer nodes which are passed
between processes. Try multi-process mode with much less scene
graph data and see if your experience is different (i.e. no crash?)
We also tend to build and test stuff in
PFMP_APP_CULL_DRAW mode even on single cpu machines because in
the end we run them on Onyx multi-cpu systems.
Thomas Hudson wrote:
>
> Morten writes:
> > Maybe the best strategy is to run the application in multiprocessing
> > mode even when I'm developing on an Indy? There always seems to be a
> > few hours of debugging to do whenever I get the chance to test out the
> > code on the quadruple CPU Onyx. Heh.
>
> I've had our Indigo2s die a really nasty death under me, reproducably, when
> I told it to use pfMultiprocess(PFMP_APP_CULL_DRAW) in 2.0. Tries to dump
> 400+ MB of core and the parent process of my window manager goes away...
>
> My sysadmin only said "Yes, if you hit your head with the hammer, IT WILL HURT"
-Randy
-- // Randy Stiles stiles++at++aic.lockheed.com Orgn 9620 Bldg 255 // Lockheed Martin Advanced Technology Center 3251 Hanover Street // office: 415.354.5256 fax: 415.354.5235 Palo Alto, CA 94304-1192 // http://vet.parl.com/~vet/people/stiles/
This archive was generated by hypermail 2.0b2 on Mon Aug 10 1998 - 17:52:36 PDT