Re: Re: Seeing 800us of latency with cyclictest on a PC with 2.6.26.8 rt patch. Is that normal?

Re: Re: Seeing 800us of latency with cyclictest on a PC with 2.6.26.8 rt patch. Is that normal?

[Matthias Luescher]

> Specifically, my laptop is a Dell Latitude D620.

We have tried very hard to achieve real time performance with the 
Dell Latitude D620 (we have tested different kernels, different configurations,
different bios versions, modified the bios settings, called the Dell customer service, ...). 
After all it looks like there is no way to achieve hard real time performance with this
type of device.

We had more luck (sub 100us latency) with the following notebooks:
- Lenovo T61, with dual core processor T8300, 2 GB RAM
- Asus V1SSeries with dual core processor T7500

On both devices we had to disable some power management features.
Please note that pressing some function keys can still result in huge latencies if 
the bios implementation is bad enough.

Desktop PCs usually work without big latencies.

Best regards
Matthias

Re: Seeing 800us of latency with cyclictest on a PC with 2.6.26.8 rt patch. Is that normal?

[Brian Hutchinson]

Matthias Luescher <m.luescher <at> vtxmail.ch> writes:

> 
> > Specifically, my laptop is a Dell Latitude D620.
> 
> We have tried very hard to achieve real time performance with the 
> Dell Latitude D620 (we have tested different kernels, different configurations,
> different bios versions, modified the bios settings, called the Dell customer
service, ...). 
> After all it looks like there is no way to achieve hard real time performance
with this
> type of device.
> 
> We had more luck (sub 100us latency) with the following notebooks:
> - Lenovo T61, with dual core processor T8300, 2 GB RAM
> - Asus V1SSeries with dual core processor T7500
> 
> On both devices we had to disable some power management features.
> Please note that pressing some function keys can still result in huge
latencies if 
> the bios implementation is bad enough.
> 
> Desktop PCs usually work without big latencies.
> 
> Best regards
> Matthias
> 
> --
> To unsubscribe from this list: send the line "unsubscribe linux-rt-users" in
> the body of a message to majordomo <at> vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
> 
> 

Just wanted to post an update.  I build 2.6.24.7 with rt patch and installed on
a 10 year old ASUS MEW motherboard with a 366MHz Celeron running Debian 5.0.

Cyclictest -a -t -n -p99 reports avg of 29us and max of 74us!!!  Wow!

OK, now I just need to do the same thing on a ARM 926EJ-S board (once I get one).

Thanks to everyone for all the suggestions.  I wouldn't have thought that 10
year old desktop hardware could beat the pants off a modern laptop.

Re: Seeing 800us of latency with cyclictest on a PC with 2.6.26.8 rt patch. Is that normal?

[Clark Williams]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

On Wed, 11 Mar 2009 19:31:01 +0000 (UTC)
Brian Hutchinson <b.hutchman@gmail.com> wrote:
> 
> Thanks to everyone for all the suggestions.  I wouldn't have thought that 10
> year old desktop hardware could beat the pants off a modern laptop.
> 
>

Almost certainly it's because your old desktop doesn't have lots of
things going on behind-the-scenes in the BIOS. Modern x86/x86_64 h/w
vendors have added lots of outside the OS functions such as thermal
monitoring and special function keys (ala the Thinklight on most
Thinkpads) that are handled via System Management Interrupts (SMI)
directly by the BIOS. If the SMI service routines take long times to do
whatever they're doing, you get unexplained latencies. 

Jon Masters wrote an SMI detector module that we've been using to good
effect to detect and classify SMIs (/me rummages for LKML link).
Actually, I think we've added some stuff since he posted it last, so
I'll poke him to post it again. It's a special purpose module that runs
a function in the stop_machine() mode. The function just polls the TSC
looking for gaps that exceed a certain threshold. Any gap in the TSC
values on a particular core are very good candidates for SMIs.

Clark
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v2.0.9 (GNU/Linux)

iEYEARECAAYFAkm4/voACgkQHyuj/+TTEp1/xwCgwgY5MprNyQ6Zt1WTMvfNP0sW
jwMAoKTq+aijqZlNA4MzN2vKl/QFf8hD
=asgJ
-----END PGP SIGNATURE-----