[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]

Re: [Linux-cluster] GFS performance.



Are you mounting with noatime parameter? That's the only thing I've found
that makes any significant difference.

4x slowdown may be on the slow side for a single node, but it's in the
right ball park. It's not going to get close to ext3 in terms of
performance. Also expect a further slow-down of a few factors as you up the
number of nodes for concurrent small random reads case.

In general - a clustered shared disk fs will never be as fast as a
non-shared file system. There is no way around the fact that it has to do
more work to arbitrate the data access between the nodes.

Gordan

On Mon, 20 Apr 2009 23:22:41 +0800, Vikash Khatuwala
<vikash netvigator com>
wrote:
> Hello,
> 
> OS : CentOS 5.2
> FS : GFS
> Journals : 4
> Nodes : 1 (currently testing)
> iSCSI Target : Dell MD3000i
> Disks : 5 x SAS 15K RPM 300GB.
> 
> I would like to know what is the expected performance penalty for using
> GFS.
> 
> Currently I have a single node cluster for testing using the lock_dlm 
> over an iSCSI RAID 5 disk system. Ive create 2 volumes on the same 
> RAID 5 disk set 1st one is GFS and the second one is ext3. Ive used 
> "fio" for my performance testing and found that ext3 is about 4 times 
> faster then GFS. The test is using random reads of 4KB. If instead I 
> use sequential reads of 256K then the performance is very close to 
> the ext3 filesystem, but thats far from our practical environment.
> 
> This seems to be quite extreme and so would like to know how to tune 
> the performance on GFS. Ive tried guidelines from:
>
http://www.redhat.com/promo/summit/2008/downloads/pdf/Thursday/Summit08presentation_GFSBestPractices_Final.pdf
> 
> However they only improve CPU utilization but disk IO performance 
> does not improve by much.
> 
> Now I need to make a decision to go with GFS or not, clearly at 4 
> times less performance we cannot afford it, also it doesn't sound 
> right so would like to find out whats wrong.
> 
> Thanks and hope to get some pointers.
> 
> Regards,
> Vikash.
> 
> --
> Linux-cluster mailing list
> Linux-cluster redhat com
> https://www.redhat.com/mailman/listinfo/linux-cluster


[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]