SCIENTIFIC-LINUX-USERS Archives

May 2011

SCIENTIFIC-LINUX-USERS@LISTSERV.FNAL.GOV

Options: Use Monospaced Font
Show Text Part by Default
Show All Mail Headers

Message: [<< First] [< Prev] [Next >] [Last >>]
Topic: [<< First] [< Prev] [Next >] [Last >>]
Author: [<< First] [< Prev] [Next >] [Last >>]

Print Reply
Subject:
From:
Reply To:
Date:
Sat, 14 May 2011 00:44:49 -0700
Content-Type:
text/plain
Parts/Attachments:
text/plain (66 lines)
Mat, have you ever measured the latency for moving data from CPU to
CPU both within a package and between packages? My partner is having
a difficult time with a very complex audio processing/mixing/delaying/
and equalizing application that can defeat any attempts to keep audio
paths on one CPU or even one package. I'm wondering what Linux RT can
achieve in this regard. We suspect the problem is in the chipsets
more than the OS. But it would be fun to learn different.

Thanks
{^_^}  Joanne

On 2011/05/13 03:21, Mathew Maher wrote:
> Hello Charles,
>
> MRG is a redhat product for Messaging/Realtime/Grid. They've packaged all this together for use with financial applications, but its only the realtime bit which has most use for scientific applications (possibly messaging).
>
> SL5 distros are here:
> http://glitesoft.cern.ch/cern/mrg/slc5X/i386/RPMS/repoview/mrg-realtime.group.html
>
> We manually patched CentOS 5 with the realtime kernel (i.e. without this RPM) and did some speed trials. Generally, it works very well, although not 100% deterministic/guaranteed.
>
> The trial machine was a dual Xeon X5550 2.66GHz (8 CPU cores). Without the RT kernel, we averaged 1ms thread wakeup/latency. With the RT kernel, this dropped to typical 5uS latency, but with occasional 'spikes' up to 60uS. I suspect these spikes are due to the processors NMI/SSI interrupts kicking in, which there is little the kernel can do about.
>
> The real killer test was running again with Hyperthreading enabled. This gives a total 16 virtual CPU's. Without RT, latency was 1.04mS. With RT this dropped to 8uS average, but spiking up to 134uS.
>
> Mat
>
> -----Original Message-----
> From: Charles ELSAESSER WebmailOrange [mailto:[log in to unmask]]
> Sent: 13 May 2011 11:04
> To: Mathew Maher
> Cc: [log in to unmask]
> Subject: re: MRG / Realtime with SL6
>
> Bonjour,
>
> It would have been nice and usefull
> if you had given links to packages, docs, patches for kernel-rt (et al) packages available for SLXXX if you know them.
>
> What does MRG stands for?
>
> Thank you.
>
> Charles
>
> --------------
>> Message du 13/05/11 10:20
>> De : "Mathew Maher"
>> A : [log in to unmask]
>> Copie à :
>> Objet : MRG / Realtime with SL6
>>
>> Hi Guys,
>> Has anyone attempted to patch SL6 into realtime yet?
>>
>> I'm specifically referring to the kernel-rt (et al) packages available
>> ...
>
> This e-mail and any attachments may contain confidential and
> privileged information. If you are not the intended recipient,
> please notify the sender immediately by return e-mail, delete this
> e-mail and destroy any copies. Any dissemination or use of this
> information by a person other than the intended recipient is
> unauthorized and may be illegal.
>

ATOM RSS1 RSS2