[Linux-cluster] GFS: drop_count and drop_period tuning
Michael.Hagmann at hilti.com
Wed Sep 12 19:50:20 UTC 2007
the Problem is that ( befor glock_purge Parameter ) no real mechanism to release glocks exists, the only limit is the memory size. Because we have a lot of Memory ( min. 32 GB RAM ) and 6 Nodes, DLM cam on its limit to handle the locks ( over 6 million ) and timed out !
That means for you, maybe you use less memory but was is more important "performance" The DLM has less glocks to handle and is faster! In our Case the Cluster was, wihout this parameter not able to run!
But I don't now how this impact the drop_cout value.
UNIX Systems Engineering
Enterprise Systems Technology
9494 Schaan Liechtenstein
Feldkircherstrasse 100 P.O.Box 333
P +423-234 2467 F +423-234 6467
E michael.hagmann at hilti.com
From: linux-cluster-bounces at redhat.com on behalf of Claudio Tassini
Sent: Tue 9/11/2007 10:35
To: linux clustering
Subject: Re: [Linux-cluster] GFS: drop_count and drop_period tuning
Thanks Michael, I've set this option on my filesystems. How should this impact to the system performance/behaviour? More/less memory usage? I guess that, by trimming the 50% of unused locks every 5 secs, it should cut off memory usage too.. am I right?
If this works, I could also raise the drop_count value?
2007/9/10, Hagmann, Michael < Michael.Hagmann at hilti.com <mailto:Michael.Hagmann at hilti.com> >:
When you are on RHEL4.5 then I highly suggest you to use the new glock_purge Parameter for every gfs Filesystem add to /etc/rc.local
gfs_tool settune / glock_purge 50
gfs_tool settune /scratch glock_purge 50
also this Parameter has to set new on every mount. That mean when you umount it and then mount it again, run the /etc/rc.local again, otherway the parameter are gone!
maybe also checkout this page --> http://www.open-sharedroot.org/Members/marc/blog/blog-on-gfs/glock-trimming-patch <http://www.open-sharedroot.org/Members/marc/blog/blog-on-gfs/glock-trimming-patch>
From: linux-cluster-bounces at redhat.com [mailto:linux-cluster-bounces at redhat <mailto:linux-cluster-bounces at redhat.com> .com <mailto:linux-cluster-bounces at redhat.com> ] On Behalf Of Claudio Tassini
Sent: Montag, 10. September 2007 13:19
To: linux clustering
Subject: [Linux-cluster] GFS: drop_count and drop_period tuning
I have a four-nodes GFS cluster on RH 4.5 (last versions, updated yesterday). There are three GFS filesystems ( 1 TB, 450 GB and 5GB), serving some mail domains with postfix/courier imap in a "maildir" configuration.
As you can suspect, this is not exactly the best for GFS: we have a lot (thousands) of very small files (emails) in a very lot of directories. I'm trying to tune up things to reach the best performance. I found that tuning the drop_count parameter in /proc/cluster/lock_dlm/drop_period , setting it to a very large value (it was 500000 and now, after a memory upgrade, I've set it to 1500000 ), uses a lot of memory (about 10GB out of 16 that I've installed in every machine) and seems to "boost" performance limiting the iowait CPU usage.
The bad thing is that when I umount a filesystem, it must clean up all that locks (I think), and sometimes it causes problems to the whole cluster, with the other nodes that stop writes to the filesystem while I'm umounting on one node only.
Is this normal? How can I tune this to clean memory faster when I umount the FS? I've read something about setting more gfs_glockd daemons per fs with the num_glockd mount option, but it seems to be quite deprecated because it shouldn't be necessary..
Linux-cluster mailing list
Linux-cluster at redhat.com
https://www.redhat.com/mailman <https://www.redhat.com/mailman/listinfo/linux-cluster> /listinfo/linux-cluster <https://www.redhat.com/mailman/listinfo/linux-cluster>
-------------- next part --------------
A non-text attachment was scrubbed...
Size: 4879 bytes
Desc: not available
More information about the Linux-cluster