Home » Mailing list archives » linux-kernel » 2008 » September » 8

Re: 2.6.27-rc5 OLTP performance regression

Previous thread: [git pull] x86 fixes by H. Peter Anvin on Monday, September 8, 2008 - 1:52 pm. (29 messages)

Next thread: [git pull] x86 fixes by H. Peter Anvin on Monday, September 8, 2008 - 3:32 pm. (15 messages)
[view original message]
From: Peter Zijlstra <peterz@...>
To: Ma, Chinang <chinang.ma@...>
Cc: Ingo Molnar <mingo@...>, Srivatsa Vaddagiri <vatsa@...>, Mike Galbraith <efault@...>, Gregory Haskins <ghaskins@...>, Steven Rostedt <rostedt@...>, Nick Piggin <nickpiggin@...>, Siddha, Suresh B <suresh.b.siddha@...>, Wilcox, Matthew R <matthew.r.wilcox@...>, Tripathi, Sharad C <sharad.c.tripathi@...>, Chilukuri, Harita <harita.chilukuri@...>, linux-kernel@vger.kernel.org <linux-kernel@...>
Subject: Re: 2.6.27-rc5 OLTP performance regression
Date: Monday, September 8, 2008 - 2:00 pm

Because the .26 group scheduler wasn't SMP aware. The extra cost comes

from the fact that .27 is.


--
[view original message]
From: Peter Zijlstra <peterz@...>
To: Ma, Chinang <chinang.ma@...>
Cc: Ingo Molnar <mingo@...>, Srivatsa Vaddagiri <vatsa@...>, Mike Galbraith <efault@...>, Gregory Haskins <ghaskins@...>, Steven Rostedt <rostedt@...>, Nick Piggin <nickpiggin@...>, Siddha, Suresh B <suresh.b.siddha@...>, Wilcox, Matthew R <matthew.r.wilcox@...>, Tripathi, Sharad C <sharad.c.tripathi@...>, Chilukuri, Harita <harita.chilukuri@...>, linux-kernel@vger.kernel.org <linux-kernel@...>
Subject: Re: 2.6.27-rc5 OLTP performance regression
Date: Monday, September 8, 2008 - 2:04 pm

What you can do it increase the /proc/sys/kernel/sched_shares_ratelimit

value and thereby decrease the accuracy of the SMP fairness of the group

scheduler.


--
[view original message]
From: Ma, Chinang <chinang.ma@...>
To: Peter Zijlstra <peterz@...>
Cc: Ingo Molnar <mingo@...>, Srivatsa Vaddagiri <vatsa@...>, Mike Galbraith <efault@...>, Gregory Haskins <ghaskins@...>, Steven Rostedt <rostedt@...>, Nick Piggin <nickpiggin@...>, Siddha, Suresh B <suresh.b.siddha@...>, Wilcox, Matthew R <matthew.r.wilcox@...>, Tripathi, Sharad C <sharad.c.tripathi@...>, Chilukuri, Harita <harita.chilukuri@...>, linux-kernel@vger.kernel.org <linux-kernel@...>
Subject: RE: 2.6.27-rc5 OLTP performance regression
Date: Tuesday, September 9, 2008 - 5:39 pm

Peter,

        We increased sched_shares_ratelimit and were able to recover the 2% performance regression.

Thanks,


-----Original Message-----

From: Peter Zijlstra [mailto:peterz@infradead.org]

Sent: Monday, September 08, 2008 11:04 AM

To: Ma, Chinang

Cc: Ingo Molnar; Srivatsa Vaddagiri; Mike Galbraith; Gregory Haskins; Steven Rostedt; Nick Piggin; Siddha, Suresh B; Wilcox, Matthew R; Tripathi, Sharad C; Chilukuri, Harita; linux-kernel@vger.kernel.org

Subject: Re: 2.6.27-rc5 OLTP performance regression


What you can do it increase the /proc/sys/kernel/sched_shares_ratelimit

value and thereby decrease the accuracy of the SMP fairness of the group

scheduler.


--
[view original message]
From: Ingo Molnar <mingo@...>
To: Ma, Chinang <chinang.ma@...>
Cc: Peter Zijlstra <peterz@...>, Srivatsa Vaddagiri <vatsa@...>, Mike Galbraith <efault@...>, Gregory Haskins <ghaskins@...>, Steven Rostedt <rostedt@...>, Nick Piggin <nickpiggin@...>, Siddha, Suresh B <suresh.b.siddha@...>, Wilcox, Matthew R <matthew.r.wilcox@...>, Tripathi, Sharad C <sharad.c.tripathi@...>, Chilukuri, Harita <harita.chilukuri@...>, linux-kernel@vger.kernel.org <linux-kernel@...>
Subject: Re: 2.6.27-rc5 OLTP performance regression
Date: Wednesday, September 10, 2008 - 8:26 am

great. Could you please post a patch for that?


	Ingo

--
[view original message]
From: Ma, Chinang <chinang.ma@...>
To: Ingo Molnar <mingo@...>
Cc: Peter Zijlstra <peterz@...>, Srivatsa Vaddagiri <vatsa@...>, Mike Galbraith <efault@...>, Gregory Haskins <ghaskins@...>, Steven Rostedt <rostedt@...>, Nick Piggin <nickpiggin@...>, Siddha, Suresh B <suresh.b.siddha@...>, Wilcox, Matthew R <matthew.r.wilcox@...>, Tripathi, Sharad C <sharad.c.tripathi@...>, Chilukuri, Harita <harita.chilukuri@...>, linux-kernel@vger.kernel.org <linux-kernel@...>
Subject: RE: 2.6.27-rc5 OLTP performance regression
Date: Wednesday, September 10, 2008 - 12:14 pm

I increased the /proc/sys/kernel/sched_shares_ratelimit from 1 millisecond to one second to reduce the frequency of shares update. That was okay for OLTP workload. As Peter mentioned, a larger value will reduce the accuracy of SMP fairness for group scheduler and the ratelimit and using a larger default value may not be good for general case.


Thanks,

Chinang

-----Original Message-----

From: Ingo Molnar [mailto:mingo@elte.hu]

Sent: Wednesday, September 10, 2008 5:26 AM

To: Ma, Chinang

Cc: Peter Zijlstra; Srivatsa Vaddagiri; Mike Galbraith; Gregory Haskins; Steven Rostedt; Nick Piggin; Siddha, Suresh B; Wilcox, Matthew R; Tripathi, Sharad C; Chilukuri, Harita; linux-kernel@vger.kernel.org

Subject: Re: 2.6.27-rc5 OLTP performance regression


great. Could you please post a patch for that?


        Ingo

--
[view original message]
From: Ingo Molnar <mingo@...>
To: Ma, Chinang <chinang.ma@...>
Cc: Peter Zijlstra <peterz@...>, Srivatsa Vaddagiri <vatsa@...>, Mike Galbraith <efault@...>, Gregory Haskins <ghaskins@...>, Steven Rostedt <rostedt@...>, Nick Piggin <nickpiggin@...>, Siddha, Suresh B <suresh.b.siddha@...>, Wilcox, Matthew R <matthew.r.wilcox@...>, Tripathi, Sharad C <sharad.c.tripathi@...>, Chilukuri, Harita <harita.chilukuri@...>, linux-kernel@vger.kernel.org <linux-kernel@...>
Subject: Re: 2.6.27-rc5 OLTP performance regression
Date: Wednesday, September 10, 2008 - 1:34 pm

does 10 msecs already get you most of the 2%? 1000 millisecond sounds

very (and probably unacceptable) coarse.


	Ingo

--
[view original message]
From: Ma, Chinang <chinang.ma@...>
To: Ingo Molnar <mingo@...>
Cc: Peter Zijlstra <peterz@...>, Srivatsa Vaddagiri <vatsa@...>, Mike Galbraith <efault@...>, Gregory Haskins <ghaskins@...>, Steven Rostedt <rostedt@...>, Nick Piggin <nickpiggin@...>, Siddha, Suresh B <suresh.b.siddha@...>, Wilcox, Matthew R <matthew.r.wilcox@...>, Tripathi, Sharad C <sharad.c.tripathi@...>, Chilukuri, Harita <harita.chilukuri@...>, linux-kernel@vger.kernel.org <linux-kernel@...>
Subject: RE: 2.6.27-rc5 OLTP performance regression
Date: Wednesday, September 10, 2008 - 8:07 pm

When set to 10 ms the regression is about 0.4%. We get most of the 2%.


Chinang

-----Original Message-----

From: Ingo Molnar [mailto:mingo@elte.hu]

Sent: Wednesday, September 10, 2008 10:35 AM

To: Ma, Chinang

Cc: Peter Zijlstra; Srivatsa Vaddagiri; Mike Galbraith; Gregory Haskins; Steven Rostedt; Nick Piggin; Siddha, Suresh B; Wilcox, Matthew R; Tripathi, Sharad C; Chilukuri, Harita; linux-kernel@vger.kernel.org

Subject: Re: 2.6.27-rc5 OLTP performance regression


does 10 msecs already get you most of the 2%? 1000 millisecond sounds

very (and probably unacceptable) coarse.


        Ingo

--
Previous thread: [git pull] x86 fixes by H. Peter Anvin on Monday, September 8, 2008 - 1:52 pm. (29 messages)

Next thread: [git pull] x86 fixes by H. Peter Anvin on Monday, September 8, 2008 - 3:32 pm. (15 messages)