Re: [PATCH 2/2] Add stats to monitor soft_limit reclaim

[Ying Han <yinghan@google.com>]

On Thu, Apr 28, 2011 at 4:51 PM, Hiroyuki Kamezawa
<kamezawa.hiroyuki@gmail.com> wrote:
> 2011/4/29 Ying Han <yinghan@google.com>:
>> On Thu, Apr 28, 2011 at 3:37 PM, Ying Han <yinghan@google.com> wrote:
>>> This patch extend the soft_limit reclaim stats to both global background
>>> reclaim and global direct reclaim.
>>>
>>> We have a thread discussing the naming of some of the stats. Both
>>> KAMEZAWA and Johannes posted the proposals. The following stats are based
>>> on what i had before that thread. I will make the corresponding change on
>>> the next post when we make decision.
>>>
>>> $cat /dev/cgroup/memory/A/memory.stat
>>> kswapd_soft_steal 1053626
>>> kswapd_soft_scan 1053693
>>> direct_soft_steal 1481810
>>> direct_soft_scan 1481996
>>>
>>> Signed-off-by: Ying Han <yinghan@google.com>
>>> ---
>>>  Documentation/cgroups/memory.txt |   10 ++++-
>>>  mm/memcontrol.c                  |   68 ++++++++++++++++++++++++++++----------
>>>  2 files changed, 58 insertions(+), 20 deletions(-)
>>>
>>> diff --git a/Documentation/cgroups/memory.txt b/Documentation/cgroups/memory.txt
>>> index 0c40dab..fedc107 100644
>>> --- a/Documentation/cgroups/memory.txt
>>> +++ b/Documentation/cgroups/memory.txt
>>> @@ -387,8 +387,14 @@ pgpgout            - # of pages paged out (equivalent to # of uncharging events).
>>>  swap           - # of bytes of swap usage
>>>  pgfault                - # of page faults.
>>>  pgmajfault     - # of major page faults.
>>> -soft_steal     - # of pages reclaimed from global hierarchical reclaim
>>> -soft_scan      - # of pages scanned from global hierarchical reclaim
>>> +soft_kswapd_steal- # of pages reclaimed in global hierarchical reclaim from
>>> +               background reclaim
>>> +soft_kswapd_scan - # of pages scanned in global hierarchical reclaim from
>>> +               background reclaim
>>> +soft_direct_steal- # of pages reclaimed in global hierarchical reclaim from
>>> +               direct reclaim
>>> +soft_direct_scan- # of pages scanned in global hierarchical reclaim from
>>> +               direct reclaim
>
> Thank you for CC.
>
> I don't have strong opinion but once we add interfaces to mainline,
> it's hard to rename them. So, it's better to make a list of what name
> we'll need in future.
>
> Now, your naming has a format as [Reason]-[Who reclaim]-[What count?]
> soft_kswapd_steal
> soft_kswapd_scan
> soft_direct_steal
> soft_direct_scan
>
> Ok, we can make a name for wmark and limit reclaim as
>
> limit_direct_steal/scan
> wmark_bg_steal/scan
>
> Then, assume we finally do round-robin scan of memcg regardless of softlimit by
> removing global LRU, what name do we have ? Hmm,
>
> kernel_kswapd_scan/steal
> kernel_direct_scan/steal
>
> ?

Johannes has the proposal to separate out reclaims on the memcg
internally and externally. And then apply the format
[Reason]-[Who reclaim]-[What count?], also i added the 4th item .

1. when the memcg hits its hard_limit
> limit_direct_steal
> limit_direct_scan

2. when the memcg hits its wmark
> wmark_kswapd_steal
> wmark_kswapd_scan

3. the global direct reclaim triggers soft_limit pushback
> soft_direct_steal
> soft_direct_scan

4. hierarchy-triggered direct reclaim
> limit_hierarchy_steal
> limit_hierarchy_scan

5. the global bg reclaim triggers soft_limit pushback
> soft_kswapd_steal
> soft_kswapd_scan

For both soft_limit reclaim and per-memcg bg reclaim, we don't have
hierarchical reclaim. Only hitting the hard_limit of parent will
trigger direct reclaim hierarchically on the children.


> BTW, your changelog has different name of counters. please fix.

sure, will fix that.

>
> And I'm sorry I'll not be very active for a while.
np.

--Ying

>
> Thanks,
> -Kame
>

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Fight unfair telecom internet charges in Canada: sign http://stopthemeter.ca/
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>