[Ocfs2-users] Ofcs2 Questions!

[Sunil Mushran]

# grep "Resources" /sys/kernel/debug/o2dlm/UUID/dlm_state
Lock Resources: 20713 (1139459)

The first number is number of live resources. Second the number
of resources created for the life of the domain.

If you see the the number of resources levelling off but the second
number incrementing, that _could_ mean that you could benefit with
a larger cache. But you have to supplement that number with the
workload. Say you are deleting and creating files. In that case, you
will see the same number movement but not because of the cache limit.

We are in the process of adding more counters to provide more visibility
to the users.

Well, using any software without support has its risks.

Using noatime will help. Increasing commit will help if workload if largely
from one node only.

On 02/02/2011 05:10 PM, Henrique Fernandes wrote:
> centos 5.5
> ocfs2 1.4
> The mail solution, dovecot+postifx+mailscaner  uses about 1 gb the machine has 1.5 in my tests i am rising it to 3gb so it have 2gb of cached
>
> How many memory do you thing i should separeted for cache ?
>
> When we made some tests  ocfs2 1.4 had better performance than 1.6 but or tests were very simple, an script that writes lots of files and anothe rone that reads it.
>
> And it is a problem to have ocfs2 1.6 in ths centos!!
>
> Should i use ocfs2 in production ?
>
> how about the commit and noatime configs ??
>
>
> thanks!!
>
> []'sf.rique
>
>
> On Wed, Feb 2, 2011 at 9:27 PM, Sunil Mushran <sunil.mushran at oracle.com <mailto:sunil.mushran at oracle.com>> wrote:
>
>     version? distro?
>
>     This workload will benefit a lot with the indexed directories available in
>     ocfs2 1.6 (and mainline and sles11).
>
>     The other thing to check is the amount memory in the virtual machines.
>     File systems need memory to cache the inodes. If memory is lacking,
>     the inodes are freed and have to be re-read from disk time and again.
>     While this is a problem even in a local fs, it is a bigger problem in a cfs
>     as a cfs needs to do lock mastery for the same inode time and again.
>
>
>     On 02/02/2011 01:09 PM, Henrique Fernandes wrote:
>>     Hello,
>>
>>     First of all, i am new at the list and i have several questions about ocfs2 performance.
>>
>>     Where i am working i am having huge performance problens with ocfs2.
>>
>>     Let me tell my envoriment.
>>
>>     3 Xen VirtualMachines withs ocfs2 mounting an LUN exported over iSCSI. ( acctualy 3 LUNS, 3 ocfs2 clusters )
>>
>>     I am not the one who configured the envoriment, but it is making the performance of my MAIL system to bad.
>>
>>     Have about 9k accounts but only 4k are active. It is a maildir system. ( postfix + dovecot )
>>
>>     Now that this performance problens are afecting my system i am gonna try help to tunning the ocfs2.
>>
>>     Pretty much all default settings.
>>
>>     OCFS2 is configured to write with ordered mode. We know that changing to writeback will make performance much better, but we are not considering lose anydata, so it i snot an option.
>>
>>     Now we are going to implemente noatime options in mount. This make better performace ?
>>
>>     Other one, how about the commit mount options ? The default is set to 5s  if i increse it how is the potential data loss in case we lost lose power?
>>
>>     Does anyone have any other paramenter that should help us ?
>>
>>     Another info, the inscremental backup is taking 10 to 12 hours.
>>
>>     All nodes have VERY high I/O wait.
>>
>>     Thanks to all!!
>>
>>     If you could tell me any doc that i sould read would be nice to!
>>
>>
>>
>>
>>
>>
>>     []'sf.rique
>>
>>
>>     _______________________________________________
>>     Ocfs2-users mailing list
>>     Ocfs2-users at oss.oracle.com  <mailto:Ocfs2-users at oss.oracle.com>
>>     http://oss.oracle.com/mailman/listinfo/ocfs2-users
>
>

-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://oss.oracle.com/pipermail/ocfs2-users/attachments/20110202/7d677b86/attachment-0001.html