2004-11-04 21:14:27

by Anton Blanchard

[permalink] [raw]
Subject: cache_hot_time


Hi,

Im catching up on all the scheduler changes, and I noticed some large
changes in cache_hot_time. All but ia64 seem to have shifted by 1000. Is
this intententional?

Anton

include/linux/topology.h: .cache_hot_time = (5*1000/2),
include/asm-i386/topology.h: .cache_hot_time = (10*1000),
include/asm-ppc64/topology.h: .cache_hot_time = (10*1000),
include/asm-x86_64/topology.h: .cache_hot_time = (10*1000),
include/asm-ia64/topology.h: .cache_hot_time = (10*1000000),
include/asm-ia64/topology.h: .cache_hot_time = (10*1000000),


2004-11-05 01:33:07

by Nick Piggin

[permalink] [raw]
Subject: Re: cache_hot_time

Anton Blanchard wrote:
> Hi,
>
> Im catching up on all the scheduler changes, and I noticed some large
> changes in cache_hot_time. All but ia64 seem to have shifted by 1000. Is
> this intententional?
>

Don't think so. They should be all in units of sched_clock()
(ie. ns), so 10ms and 2.5ms are surely the intended values here.

> Anton
>
> include/linux/topology.h: .cache_hot_time = (5*1000/2),
> include/asm-i386/topology.h: .cache_hot_time = (10*1000),
> include/asm-ppc64/topology.h: .cache_hot_time = (10*1000),
> include/asm-x86_64/topology.h: .cache_hot_time = (10*1000),
> include/asm-ia64/topology.h: .cache_hot_time = (10*1000000),
> include/asm-ia64/topology.h: .cache_hot_time = (10*1000000),
>

2004-11-05 05:08:04

by Anton Blanchard

[permalink] [raw]
Subject: [PATCH] reset cache_hot_time


> Don't think so. They should be all in units of sched_clock()
> (ie. ns), so 10ms and 2.5ms are surely the intended values here.

OK how does this look?

Anton

--

Reset cache_hot_time to sane values (in the ms range). Some recent
changes resulted in values in the us range.

Signed-off-by: Anton Blanchard <[email protected]>

===== include/asm-i386/topology.h 1.12 vs edited =====
--- 1.12/include/asm-i386/topology.h 2004-10-19 15:26:52 +10:00
+++ edited/include/asm-i386/topology.h 2004-11-05 15:50:13 +11:00
@@ -78,7 +78,7 @@
.max_interval = 32, \
.busy_factor = 32, \
.imbalance_pct = 125, \
- .cache_hot_time = (10*1000), \
+ .cache_hot_time = (10*1000000), \
.cache_nice_tries = 1, \
.per_cpu_gain = 100, \
.flags = SD_LOAD_BALANCE \
===== include/asm-ppc64/topology.h 1.13 vs edited =====
--- 1.13/include/asm-ppc64/topology.h 2004-10-19 15:26:52 +10:00
+++ edited/include/asm-ppc64/topology.h 2004-11-05 15:49:52 +11:00
@@ -46,7 +46,7 @@
.max_interval = 32, \
.busy_factor = 32, \
.imbalance_pct = 125, \
- .cache_hot_time = (10*1000), \
+ .cache_hot_time = (10*1000000), \
.cache_nice_tries = 1, \
.per_cpu_gain = 100, \
.flags = SD_LOAD_BALANCE \
===== include/asm-x86_64/topology.h 1.13 vs edited =====
--- 1.13/include/asm-x86_64/topology.h 2004-10-19 15:26:52 +10:00
+++ edited/include/asm-x86_64/topology.h 2004-11-05 15:49:37 +11:00
@@ -42,7 +42,7 @@
.max_interval = 32, \
.busy_factor = 32, \
.imbalance_pct = 125, \
- .cache_hot_time = (10*1000), \
+ .cache_hot_time = (10*1000000), \
.cache_nice_tries = 1, \
.per_cpu_gain = 100, \
.flags = SD_LOAD_BALANCE \
===== include/linux/topology.h 1.6 vs edited =====
--- 1.6/include/linux/topology.h 2004-10-19 15:26:51 +10:00
+++ edited/include/linux/topology.h 2004-11-05 15:48:15 +11:00
@@ -113,7 +113,7 @@
.max_interval = 4, \
.busy_factor = 64, \
.imbalance_pct = 125, \
- .cache_hot_time = (5*1000/2), \
+ .cache_hot_time = (5*1000000/2), \
.cache_nice_tries = 1, \
.per_cpu_gain = 100, \
.flags = SD_LOAD_BALANCE \

2004-11-05 08:32:50

by Nick Piggin

[permalink] [raw]
Subject: Re: [PATCH] reset cache_hot_time

Anton Blanchard wrote:
>
>
>>Don't think so. They should be all in units of sched_clock()
>>(ie. ns), so 10ms and 2.5ms are surely the intended values here.
>
>
> OK how does this look?
>

Seems like the right thing to do. Andrew's picked it up in
rc1-mm3. Thanks.