Mailstore filesystem

[Daniel Eckl]

Well, I was surprised about the problem... I have only about 250
mailboxes on that server. But it's used extensively, because some users
have over 200 folders and an incredible mail count (mailboxes over 2 GB
of size).

I assume, my problem comes from Outlooks scanning all headers of a big
folder, because the load and IO increase comes with a lot of disk
activity (high bi/bo in vmstat).

But it has to be only one or two clients, because in the main usage time
there is mostly no problem and disk io is moderate to low...

I cannot identify the triggering client and/or the action it starts....

Thanks for your suggestions, I will try if that makes a difference and
report to the list for the archives!

Thanks,
Daniel

Wesley Craig schrieb:
> On 03 Jul 2006, at 11:35, Daniel Eckl wrote:
>> At this time I didn't change the scheduler. In fact, I actually learned
>> about it a few days ago... So it was SuSE 9.3 default (might be vanilla
>> default, don't know)
> 
> I think "anticipatory" is the default IO scheduler for SuSE 9.3 (from
> Google).
> 
>> What would you recommend for cyrus? (or for ext3 running cyrus)? And are
>> there prefered scheduler for other fs, too, especially reiser and xfs?
> 
> It would probably depend on your setup.  Given the problem you describe,
> tho, I would think that "deadline" would improve your performance.  The
> 2.6 kernel has many tunable parameters.  If you're pushing a server at
> all near its limits, you probably need to become an expert in
> performance tuning.  Or buy more hardware :)
> 
> :wes