Re: Inconsistent load average on tickless kernels

["Lesław Kopeć" <leslaw.kopec@nasza-klasa.pl>]

[-- Attachment #1: Type: text/plain, Size: 1303 bytes --]

On 04/17/2012 05:30 PM, Jonathan Nieder wrote:

>> Looking at results for 2.6.32.55 branch it seems that we're back at
>> 74f5187ac8 patch - the values are almost the same. The difference
>> between NO_HZ and HZ versions is noticeable.
> 
> How does 3.0.y + c308b56b53 do?  (I ask because the usual flow of
> fixes is mainline -> 3.3.y -> 3.2.y -> 3.0.y -> 2.6.32.y with the
> first three steps happening pretty quickly, so it we can get this
> working on 3.0.y then that would be progress.  Also because, like
> 2.6.32.y, 3.0.y is longterm maintained, so it might be useful in the
> meantime.)

It seems that load reported by 3.0.28 with and without c308b56b53 patch
is almost identical to 3.2.12. This might be more clearly visible on a
comparison chart. [1]

2.6.32.55-hz-0f004f5a69			9.88
2.6.32.55-no-hz-74f5187ac8		2.48
2.6.32.55-no-hz-c308b56b53		2.22
3.0.28-hz				10.66
3.0.28-no-hz				0.60
3.0.28-no-hz-c308b56b53			4.09
3.0.28-no-hz-c308b56b53 nohz=off	6.78
3.2.12-hz				10.16
3.2.12-no-hz				0.66
3.2.12-no-hz-c308b56b53			4.36

What's worth noting is that I haven't seen any nasty side effects of the
latest patch on all kernel versions that I've tested. Hope that helps.


[1] http://img855.imageshack.us/img855/2204/kernelload.png

-- 
Lesław Kopeć


[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 262 bytes --]