You don't mention it but I assume you are running on Fiber Channel not
iSCSI? I am running GFS using CentOS 5.1 from two servers using iSCSI. I have
Device Mapper Multipath setup so that makes my performance characteristics a
little different but I have tuned the servers with the following commands:
gfs_tool settune <mountpoint> demotes_secs 600
gfs_tool settune <mountpoint> scand_secs 10
gfs_tool settune <mountpoint> statfs_slots 128
The first two I have set until 5.2 drops which will contain an additional GFS
tuning parameter, glock_purge. Update 5 of CentOS 4 may already have the
glock_purge parameter available. Run the command 'gfs_tool gettune
<mountpoint>' and see if glock_purge is listed. It should be set to
zero by default.
If it is there you can refer to this brief write up on tuning with glock_purge
since you will not want to use the demote_secs and scand_secs settings I am
using:
http://open-sharedroot.org/Members/marc/blog/blog-on-gfs/glock-trimming-patch/?searchterm=glocks
Also, as a further note, your particular usage scenario will dictate what
tunings are appropriate. I am not familiar with Oracle's usage of I/O but
will both servers be accessing the same data on disk (particularly with regards
to writing) at the same time?
-James Thompson
----- Original Message -----
From: "lingu" <hicheerup at gmail.com>
To: centos at centos.org
Sent: Monday, June 9, 2008 4:54:44 AM GMT -05:00 US/Canada Eastern
Subject: [CentOS] Slow gfs performance
HI,
?Sorry for repeating same mail ,while composing that mail i mistakenly typed the
send button.I am facing problem with my gfs and below is the? running setup.
My setup
Two node Cluster[only to create shared gfs file system] with manual fencing
running on centos 4 update 5 for oracle apps.
?Shared gfs partition are mounted on both the node[active-active]
?Whenever i type df -h command it will take some delay to print my shared gfs
partition,it is happening even if i continuously typing df -h command . Shared
GFS file system storage is from IBM STORAGE.I don't know what parameter to
be fine tuned,bcoz? of that may be i feel my? oracle apps is also very slow.
In syslog i am getting following error continuously in one of the node
less /var/log/messages
test2 kernel: 493 [RAIDarray.mpp]AHA-SN01:1:0:10 Cmnd failed-retry the same
path. vcmnd SN 7147770 pdev H2:C0:
T0:L10 0x00/0x00/0x00 0x00020000 mpp_status:2
?Any one help me to get rid of this slow gfs performance.
Regards
lingu
_______________________________________________
CentOS mailing list
CentOS at centos.org
http://lists.centos.org/mailman/listinfo/centos