Re: Graphics Glitches

New Message Reply Date view Thread view Subject view Author view

Rob Jenkins (robj++at++sgi.com)
Thu, 01 Oct 1998 11:47:17 +0100


Max, Steve, Angus etc have covered the main things to look out for and
try, I'll add a couple.

Firstly make sure you have a resoinably up to date patch set, esp the
gfx patches ( presumably if you have a support call then that's taken
care of ).

Steve mentioned doing this but some detail:

Make some CPUs non-premptive for performer to run on ( stop the system
interupts ):put NOINTR: 1 2 3 in /var/sysgen/system/irix.sm ( assuming
a 4 CPU machine - the file is well commented so have a look in there )

chkconfig off deamons you don't need, common ones are:
   % chkconfig desktop off
   % chkconfig objectserver off
   % chkconfig directoryserver off
   % chkconfig fontserver off
   % chkconfig soundscheme off

but there might be others. sockd is one that often seems to jump in but
most likely you need that, I remember so instances with early 6.4 where
it seemed to interfere with pf processes but I thin that gone away now,
with all the other steps done right then esp with 16 CPUs you should
have plenty to let the system stuff run on some OK and dedicate others
to pf.
One thing to also be sure of is running fullscreen, then you can be sure
you're not getting gfx context switches.

The *best* way to track these problems that I know is with par -rQQt<N>
where N is enough seconds to capture a glitch. This output will be big
for your machine but it will show what ran on what CPU, what it was
kicked off by and so on. You should be able to work out of some system
thing is preempting your pf procs then fix it by nri, NOINTR, locking,
isolating, restricting etc
Also, setenv PFNFYLEVEL 7 before running the pf app so you can look at
what performer managed ( or failed ) to run where.

Can you tell me the support call ref BTW ?

Cheers
Rob

John Auborn wrote:
>
> I have npri -h35 'd it and still get missed frames although not as bad.
> (I thought the -N option did that anyway).
>
> The only things I see running with top are sockd, bdflush, and the other
> normal low priority processes. Could sockd be a problem? How do I turn it off?
>
> I have also tried turning off all the networking and everything else I can
> think of.
>
> (and no, there is no clock program running!!)
>
> >
> > And you have npri -h 'd the processes, right?
> >
> > What does top say the processes are running at?
> >
> > Max.
>
> --
> John Auborn Code 471130D Naval Air Warfare Center Weapons Division
> Chinalake, CA 93555-6100
> (760) 939-2274 (office) (760) 939-0547 (lab)
> (760) 939-6893 (fax)
> auborn++at++chinalake.navy.mil
> http://circe.chinalake.navy.mil (work)
> http://www1.ridgecrest.ca.us/~auborn (home)
> =======================================================================
> List Archives, FAQ, FTP: http://www.sgi.com/Technology/Performer/
> Submissions: info-performer++at++sgi.com
> Admin. requests: info-performer-request++at++sgi.com

-- 
________________________________________________________________
Rob Jenkins	Silicon Graphics 	mailto:robj++at++sgi.com

New Message Reply Date view Thread view Subject view Author view

This archive was generated by hypermail 2.0b2 on Thu Oct 01 1998 - 03:47:25 PDT

This message has been cleansed for anti-spam protection. Replace '++at++' in any mail addresses with the '@' symbol.