* Hyperthreading performance on 2.4.19 and 2.5.32
@ 2002-09-17 20:03 Duc Vianney
2002-09-17 20:05 ` [Lse-tech] " Martin J. Bligh
2002-09-17 20:35 ` Randy.Dunlap
0 siblings, 2 replies; 5+ messages in thread
From: Duc Vianney @ 2002-09-17 20:03 UTC (permalink / raw)
To: linux-kernel, lse-tech
The following are data comparing the effects of hyperthreading (HT)on
stock kernel 2.4.19 and 2.5.32.
Hardware under test. The hardware is a Xeon 1-CPU MP, 1.6 gigahertz,
and 2.5 GB RAM.
Kernel under test. When testing under 2.4.19, the kernel was built
as an SMP kernel, and was run on the hardware with HT enabled through
the boot option 'noht'. When testing under 2.5.32, the kernel was
built as an SMP kernel, and was run on the hardware with HT enabled
through selecting ACPI in configuration.
Benchmarks. For multithreaded benchmarks: chat, dbench and tbench.
Summary of results. The results on Linux kernel 2.4.19 show HT might
improve multithreaded application by as much as 30%. On kernel 2.5.32,
HT may provide speed-up as high as 60%.
Observations. There are two major differences between 2.4.19 and
2.5.32 which could affect HT performance: O(1)scheduler and Ingo's
shared runqueue patch for HT that went in 2.5.32. However, Ingo's HT
patch is for handling load balancing, affinity, and task pickup. Those
are problems that exist in systems with >= 2CPUs. Since I have only
1-CPU in my test, I think the O(1) scheduler has had greater impact
than the runqueue patch. On 2.5.32, the chat workload seems to benefit
the most, followed by tbench and dbench.
The data for each number of chat rooms run (e.g., 20) represents the
geometric mean of five runs. Same method was also used for each number
of clients run in dbench and tbench.
chat workload 2.4.19 2.5.32
No. chat rooms Speed-up Speed-up
20 24% 51%
30 22% 41%
40 22% 60%
50 28% 39%
Geometric Mean 24% 45%
dbench workload 2.4.19 2.5.32
No.clients Speed-up Speed-up
20 29% 27%
30 29% 9%
60 12% 1%
90 9% 4%
120 16% 23%
Geometric Mean 18% 12%
tbench workload 2.4.19 2.5.32
No.clients Speed-up Speed-up
20 31% 36%
30 30% 36%
60 26% 36%
90 22% 35%
120 27% 33%
Geometric Mean 27% 35%
Duc Vianney - dvianney@us.ibm.com
^ permalink raw reply [flat|nested] 5+ messages in thread* Re: [Lse-tech] Hyperthreading performance on 2.4.19 and 2.5.32
2002-09-17 20:03 Hyperthreading performance on 2.4.19 and 2.5.32 Duc Vianney
@ 2002-09-17 20:05 ` Martin J. Bligh
2002-09-17 20:35 ` Randy.Dunlap
1 sibling, 0 replies; 5+ messages in thread
From: Martin J. Bligh @ 2002-09-17 20:05 UTC (permalink / raw)
To: Duc Vianney, linux-kernel, lse-tech
> Benchmarks. For multithreaded benchmarks: chat, dbench and tbench.
> Summary of results. The results on Linux kernel 2.4.19 show HT might
> improve multithreaded application by as much as 30%. On kernel 2.5.32,
> HT may provide speed-up as high as 60%.
What happened to the -38% degradation you found? That seems to have
fallen off the results list for some reason ... did you fix it, or is it still there?
M.
^ permalink raw reply [flat|nested] 5+ messages in thread
* Re: [Lse-tech] Hyperthreading performance on 2.4.19 and 2.5.32
2002-09-17 20:03 Hyperthreading performance on 2.4.19 and 2.5.32 Duc Vianney
2002-09-17 20:05 ` [Lse-tech] " Martin J. Bligh
@ 2002-09-17 20:35 ` Randy.Dunlap
1 sibling, 0 replies; 5+ messages in thread
From: Randy.Dunlap @ 2002-09-17 20:35 UTC (permalink / raw)
To: Duc Vianney; +Cc: linux-kernel, lse-tech
On Tue, 17 Sep 2002, Duc Vianney wrote:
| The following are data comparing the effects of hyperthreading (HT)on
| stock kernel 2.4.19 and 2.5.32.
| Hardware under test. The hardware is a Xeon 1-CPU MP, 1.6 gigahertz,
| and 2.5 GB RAM.
| Kernel under test. When testing under 2.4.19, the kernel was built
| as an SMP kernel, and was run on the hardware with HT enabled through
| the boot option 'noht'.
HT enabled with 'noht' ??
'noht' means "no HT", "no hyperthreading", disabled.
| When testing under 2.5.32, the kernel was
| built as an SMP kernel, and was run on the hardware with HT enabled
| through selecting ACPI in configuration.
--
~Randy
^ permalink raw reply [flat|nested] 5+ messages in thread
* Re: [Lse-tech] Hyperthreading performance on 2.4.19 and 2.5.32
@ 2002-09-17 20:49 Duc Vianney
0 siblings, 0 replies; 5+ messages in thread
From: Duc Vianney @ 2002-09-17 20:49 UTC (permalink / raw)
To: Martin J. Bligh; +Cc: linux-kernel, lse-tech
>What happened to the -38% degradation you found? That seems to have
>fallen off the results list for some reason ... did you fix it, or is it
still >there?
Martin ... Thanks for pointing it out.
The -38% degradation was seen on Sync Random Disk Writes, Sync
Sequential Disk Writes, and Sync Disk Copies observed from the
AIM9 bencmark running in Single User test mode. The degradation
is still there and I will investigate it later when we have the
hardware resource back.
Thanks ... Duc.
"Martin J. Bligh" <mbligh@aracnet.com>@lists.sourceforge.net on 09/17/2002
03:05:35 PM
Sent by: lse-tech-admin@lists.sourceforge.net
To: Duc Vianney/Austin/IBM@IBMUS, linux-kernel@vger.kernel.org,
lse-tech@lists.sourceforge.net
cc:
Subject: Re: [Lse-tech] Hyperthreading performance on 2.4.19 and 2.5.32
> Benchmarks. For multithreaded benchmarks: chat, dbench and tbench.
> Summary of results. The results on Linux kernel 2.4.19 show HT might
> improve multithreaded application by as much as 30%. On kernel 2.5.32,
> HT may provide speed-up as high as 60%.
What happened to the -38% degradation you found? That seems to have
fallen off the results list for some reason ... did you fix it, or is it
still there?
M.
-------------------------------------------------------
This SF.NET email is sponsored by: AMD - Your access to the experts
on Hammer Technology! Open Source & Linux Developers, register now
for the AMD Developer Symposium. Code: EX8664
http://www.developwithamd.com/developerlab
_______________________________________________
Lse-tech mailing list
Lse-tech@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/lse-tech
^ permalink raw reply [flat|nested] 5+ messages in thread* Re: [Lse-tech] Hyperthreading performance on 2.4.19 and 2.5.32
@ 2002-09-17 21:22 Duc Vianney
0 siblings, 0 replies; 5+ messages in thread
From: Duc Vianney @ 2002-09-17 21:22 UTC (permalink / raw)
To: Randy.Dunlap; +Cc: linux-kernel, lse-tech
>HT enabled with 'noht' ??
>'noht' means "no HT", "no hyperthreading", disabled.
OOPS .. typing too fast .. 'noht' means not-enabled.
HT is the default option on 2.4.19.
Duc.
"Randy.Dunlap" <rddunlap@osdl.org> on 09/17/2002 03:35:53 PM
To: Duc Vianney/Austin/IBM@IBMUS
cc: <linux-kernel@vger.kernel.org>, <lse-tech@lists.sourceforge.net>
Subject: Re: [Lse-tech] Hyperthreading performance on 2.4.19 and 2.5.32
On Tue, 17 Sep 2002, Duc Vianney wrote:
| The following are data comparing the effects of hyperthreading (HT)on
| stock kernel 2.4.19 and 2.5.32.
| Hardware under test. The hardware is a Xeon 1-CPU MP, 1.6 gigahertz,
| and 2.5 GB RAM.
| Kernel under test. When testing under 2.4.19, the kernel was built
| as an SMP kernel, and was run on the hardware with HT enabled through
| the boot option 'noht'.
HT enabled with 'noht' ??
'noht' means "no HT", "no hyperthreading", disabled.
| When testing under 2.5.32, the kernel was
| built as an SMP kernel, and was run on the hardware with HT enabled
| through selecting ACPI in configuration.
--
~Randy
^ permalink raw reply [flat|nested] 5+ messages in thread
end of thread, other threads:[~2002-09-17 21:17 UTC | newest]
Thread overview: 5+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2002-09-17 20:03 Hyperthreading performance on 2.4.19 and 2.5.32 Duc Vianney
2002-09-17 20:05 ` [Lse-tech] " Martin J. Bligh
2002-09-17 20:35 ` Randy.Dunlap
-- strict thread matches above, loose matches on Subject: below --
2002-09-17 20:49 Duc Vianney
2002-09-17 21:22 Duc Vianney
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox