Re: pfMultiPipe + pfMultiProcess

New Message Reply Date view Thread view Subject view Author view

Chris Tanner (cct++at++faith)
Mon, 20 Jun 1994 11:04:06 -0700

On Jun 20, 11:47am, Ran Yakir wrote:
> Subject: pfMultiPipe + pfMultiProcess
> Hi
>
> A problem :
>
> Lets say you have a 4 CPU, dual headed ONYX machine, and you want to run a
dual
> pipe performer application. The following pfMultiProcess modes give you the
> following results :
>
> 1. PFMP_APPCULLDRAW - runs both channels on one CPU (bad)
> 2. PFMP_DEFAULT, PFMP_APP_CULL_DRAW - runs the system with 5 processes (bad -
> because you have only 4 CPUs)
> 3. PFMP_APPCULL_DRAW - not permited, because performer won't allow both culls
> in the same process
>
> No item here gives good results. This is due to the fact that
multi-processing
> mode is the same for all pipes. If multi-processing mode was per pipe - I
could
> run one pipe as PFMP_APPCULL_DRAW, and the other as PFMP_APP_CULL_DRAW. Then
I
> would fully occupy my 4 CPUs.
> I know that 4 CPUs in not the bets configuration for running a dual pipe
> performer application, but I think that allowing a separate multi-processing
> mode for pipes would help in similar cases.
>
>
> Thanks
>
> Ran Yakir

>-- End of excerpt from Ran Yakir

Because of current complexities in the Performer MP framework, if you are using
multiple pfPipes then you must have the cull process forked (because they dont
share views and thus they dont share culls). This means that you should try
just forking the cull (APP_CULLDRAW).

That should give you (if you look at procs with ps for instance):
1) App
2) (a clock wrapper proc that doesnt eat much cpu)...
3) CullDraw
4) CullDraw

Hope that helps...

Later,
Chris Tanner
IRIS Performer

Remember, if you really want best real-time results make sure you take a look
at 'lockcpu.c' in the 1.2 /usr/src/Performer/src/lib/libpfutil directory such
that you can lock the procs down on the corresponding cpus and ensure no
competition for processors or context switches...

-- 


_____________________________________________________________
Chris Tanner (cct++at++faith.asd.sgi.com)
Silicon Graphics - Advanced Graphics Division
_____________________________________________________________

New Message Reply Date view Thread view Subject view Author view

This archive was generated by hypermail 2.0b2 on Mon Aug 10 1998 - 17:50:21 PDT

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