From: Dorosky, Christopher G (christopher.g.dorosky++at++lmco.com)
Date: 12/18/2001 11:59:57
I think you can put whatever you want there. Make sure you are superuser,
and can access the highest resolution clock.
You can assign your own priority to the pfcompute process. If you are
running dual cpu, you will be much better off putting the sync tasks on one,
and tasks like this on another. But since linux doesn't allow you to do
this, you have to be sneaky with priorities and kind of force this.
Christopher Dorosky
Lead Electronic Systems Engineer - Real Time Simulation
Lockheed Martin Missiles and Fire Control - Dallas
christopher.g.dorosky++at++lmco.com
972-603-2349
-----Original Message-----
From: Marc Mendez [mailto:mmendez++at++silicon-worlds.fr]
Sent: Tuesday, December 18, 2001 10:43 AM
To: info-performer++at++sgi.com
Subject: pfQuestion about multithreading/processing.
Hi Performers,
I ve got one more 2-cents question (if you need more, just ask, I ll
give you some 8)
First, the platform I m using is Linux RH7.2, pf2.5 (demo licenced for
the moment, might change 8))), SGI330 1CPU
I am trying to animate an object using data coming one per second from
the network.
Those datas are UTC-timestamped.
What I ve tried to do is creating a (default paramed) pthread listening
a socket and when data is present, put it in a buffer.
Another (default paramed) thread looks at that buffer, finds the data
for the current UTC and applies them.
But, I ve noticed that visual performances were bad : a lot of lost
frames and the fps is far below 20 (I need a 25 fps at least for video
projection).
One of the experimentations I led is : commenting out everything in my
two threads and just putting sched_yield() ...
The problem is still the same. It looks like this scheduler is taking ms
for context switching.
So I ve got a typical Linux performance pb ...
Has anyone an idea about how to deal with the few free CPU percents left
on my system ?
One solution might be trying some different values for the system time
slice (default = 100ms, reducing to 10ms might be benefiX). But I dont
know the influence of this on pThread that are user-level scheduled in
my case.
I ve been suggested to use the PFCOMPUTE callback. As I ve never used
it, can it suit my needs ? What kind of computation is allowed and which
one is forbidden in this function ?
My socket reading/managing is totally asynchronous and pfCompute seems
devoted to such a task, is the scheduling behaviour different from the
pthread ones ?
pfThanks,
Marc.
-----------------------------------------------------------------------
List Archives, Info, FAQ: http://www.sgi.com/software/performer/
Open Development Project: http://oss.sgi.com/projects/performer/
Submissions: info-performer++at++sgi.com
Admin. requests: info-performer-request++at++sgi.com
-----------------------------------------------------------------------
This archive was generated by hypermail 2b29 : Tue Dec 18 2001 - 12:04:40 PST