Groups > comp.os.linux.development.system > #401

Howto - or - Feature Request - Process to Core locking

Path	csiph.com!newsfeed.hal-mli.net!feeder3.hal-mli.net!newsfeed.hal-mli.net!feeder1.hal-mli.net!npeer01.iad.highwinds-media.com!news.highwinds-media.com!feed-me.highwinds-media.com!postnews.google.com!glegroupsg2000goo.googlegroups.com!not-for-mail
From	John S <john.s43851955@gmail.com>
Newsgroups	comp.os.linux.development.system
Subject	Howto - or - Feature Request - Process to Core locking
Date	Wed, 7 Mar 2012 14:00:26 -0800 (PST)
Organization	http://groups.google.com
Lines	28
Message-ID	<25803231.739.1331157626474.JavaMail.geo-discussion-forums@vbbfv2> (permalink)
NNTP-Posting-Host	24.126.95.12
Mime-Version	1.0
Content-Type	text/plain; charset=ISO-8859-1
X-Trace	posting.google.com 1331157710 373 127.0.0.1 (7 Mar 2012 22:01:50 GMT)
X-Complaints-To	groups-abuse@google.com
NNTP-Posting-Date	Wed, 7 Mar 2012 22:01:50 +0000 (UTC)
Complaints-To	groups-abuse@google.com
Injection-Info	glegroupsg2000goo.googlegroups.com; posting-host=24.126.95.12; posting-account=-NFKmwoAAAAl0IsBTNx2_eZpsoQrFUwM
User-Agent	G2/1.0
X-Received-Bytes	2329
Xref	csiph.com comp.os.linux.development.system:401

Show key headers only | View raw

  My project is currently developing a DSP application that will run
on a dual-cpu X5670 system - having a total of 12 cores/24 threads.

  We have an issue we have yet to find a solution for.  Our Signal
Processing engines are locked to specific hyperthreads (a total of 8
of the 24 hyperthreads) in an attempt to minimize conflict with other
tasking.  Each engine typically uses about 80% of that threads CPU
time.
  The problem is, during the 20% time that engine is idle, the
operating system schedule sometimes places other tasks on one of the 8
hyperthreads the engines are using - because the current cpu usage is
approx 0%.  This is a problem if the scheduler places another process
on one of the engine's core/hyperthreads while that engine is idle
because when the engine comes out of suspend, it will need the
undivided attention of that core - least it not be able to keep up
with the data stream.
   We have tried cpuset and taskset to control affinity, but these
haven't fully resolved the issues.  The CPU scheduler will still
utilize an engine's core even though other cores are either unused or
have low useage.
   Is there a way to have the CPU scheduler assign processes based on
average CPU usage?  If so, this would keep the scheduler from using an
engine's core.

   Any suggestions or ideas would be greatly appreciated.

  Thanks,
  John

Back to comp.os.linux.development.system | Previous | Next — Next in thread | Find similar

Thread

Howto - or - Feature Request - Process to Core locking John S <john.s43851955@gmail.com> - 2012-03-07 14:00 -0800
  Re: Howto - or - Feature Request - Process to Core locking "Ersek, Laszlo" <lacos@caesar.elte.hu> - 2012-03-08 01:17 +0100
    Re: Howto - or - Feature Request - Process to Core locking John S <john.s43851955@gmail.com> - 2012-03-11 03:02 -0700

csiph-web