[RFT][PATCH] sched: scale sysctl_sched_shares_ratelimit with nr_cpus

[RFT][PATCH] sched: scale sysctl_sched_shares_ratelimit with nr_cpus

[Peter Zijlstra]

[And now with proper lkml address - so much for auto-add address entries]

Dhaval, Vatsa,

Could you guys give this patch a spin on the big iron and possibly tune
the default shares_ratelimit value to give satisfactory fairness on your
large machines while considering the overhead?


---
Subject: sched: scale sysctl_sched_shares_ratelimit with nr_cpus

David reported that his Niagra spend a little too much time in
tg_shares_up(), which considering he has a large cpu count makes sense.

So scale the ratelimit value with the number of cpus like we do for
other controls as well.

Reported-by: David Miller <davem@davemloft.net>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
---
 kernel/sched.c |    6 ++++--
 1 file changed, 4 insertions(+), 2 deletions(-)

Index: linux-2.6/kernel/sched.c
===================================================================
--- linux-2.6.orig/kernel/sched.c
+++ linux-2.6/kernel/sched.c
@@ -809,9 +809,9 @@ const_debug unsigned int sysctl_sched_nr
 
 /*
  * ratelimit for updating the group shares.
- * default: 0.5ms
+ * default: 0.25ms
  */
-const_debug unsigned int sysctl_sched_shares_ratelimit = 500000;
+const_debug unsigned int sysctl_sched_shares_ratelimit = 250000;
 
 /*
  * period over which we measure -rt task cpu usage in us.
@@ -5732,6 +5732,8 @@ static inline void sched_init_granularit
 		sysctl_sched_latency = limit;
 
 	sysctl_sched_wakeup_granularity *= factor;
+
+	sysctl_sched_shares_ratelimit *= factor;
 }
 
 #ifdef CONFIG_SMP

Re: [RFT][PATCH] sched: scale sysctl_sched_shares_ratelimit with nr_cpus

[Peter Zijlstra]

On Wed, 2008-07-30 at 03:29 -0700, David Miller wrote:
> From: Peter Zijlstra <peterz@infradead.org>
> Date: Wed, 30 Jul 2008 12:25:11 +0200
> 
> > Dhaval, Vatsa,
> > 
> > Could you guys give this patch a spin on the big iron and possibly tune
> > the default shares_ratelimit value to give satisfactory fairness on your
> > large machines while considering the overhead?
> > 
> > 
> > ---
> > Subject: sched: scale sysctl_sched_shares_ratelimit with nr_cpus
> > 
> > David reported that his Niagra spend a little too much time in
> > tg_shares_up(), which considering he has a large cpu count makes sense.
> > 
> > So scale the ratelimit value with the number of cpus like we do for
> > other controls as well.
> > 
> > Reported-by: David Miller <davem@davemloft.net>
> > Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
> 
> This turned out to be a false theory, and the lockdep problem I
> reported the past few days was the real culprit.
> 
> The rebalancing et al. showing up in my debugging dumps was
> just chance and not even consistent.

Ah, still I think the proposed change makes sense, but lets wait for
some test results from Dhaval and or Vatsa.. ;-)