[Lucid] SRU: request pull EC2 Revert "sched: update load count only once per cpu in 10 tick update window"

John Johansen john.johansen at canonical.com
Fri Aug 20 18:47:58 UTC 2010


On 08/20/2010 11:34 AM, Chase Douglas wrote:
> On Fri, 2010-08-20 at 11:01 -0700, John Johansen wrote:
>> The following changes since commit 65316c3b7e1fccaae5f2c0cf5f98cd3220a5be64:
>>   Stefan Bader (1):
>>         UBUNTU: Ubuntu-2.6.32-308.15
>>
>> are available in the git repository at:
>>
>>   git://kernel.ubuntu.com/jj/ubuntu-lucid ec2
>>
>> John Johansen (1):
>>       UBUNTU: SAUCE: Revert "sched: update load count only once per cpu in 10 tick update window"
>>
>>  kernel/sched.c |   24 ++----------------------
>>  1 files changed, 2 insertions(+), 22 deletions(-)
>>
>> ------
>>
>> UBUNTU: SAUCE: Revert "sched: update load count only once per cpu in 10 tick update window"
>>
>> BugLink: http://bugs.launchpad.net/bugs/574910
>>
>> SRU Justification:
>>
>> Impact:
>> Fixes loadavg reporting on EC2.
>>
>> Fix:
>> This reverts commit 0d843425672f4d2dc99b9004409aae503ef4d39f which fixes a bug in load
>> accounting when a tickless (no idle HZ) kernel is used.  However the Xen patchset used
>> on EC2 is not tickless but the accounting modifications are still being done, resulting
>> in phantom load.
>>
>> Testcase:
>> Start any Ubuntu Lucid based instance on EC2, let it idle while logging the load average.
>>      while true ; do cat /proc/loadavg >>load.log ; sleep 5 ; done
>> Alternately simply run top or htop and monitor the load average.
>>
>> Without the revert the reported load will vary from 0 up to about .5 for a clean image
>> with no extra tasks launched.
>>
>> With the revert the load stays steady around 0 with only occasional small bump when
>> a background task is run.
> 
> On IRC, John said using the upstream version of this patch fixes the
> issue, so I'm comfortable with swapping the patches as a resolution.
> 
> I'm more against removing this patch and not putting in the upstream
> patch. However, I'll defer to John if he thinks the non-tickless Xen
> kernel doesn't need this fix and it's a better solution for ec2.
> 
Right the upstream version does fix the issue.  I specifically didn't
request the upstream version because for the EC2 kernel which is NOT
tickless as it is effectively no different than just reverting the patch,
so I chose the smaller change set.

If preferred I can reissue the SRU request with the upstream patch applied.






More information about the kernel-team mailing list