The device mappings were nothing out of the ordinary, LSI Logic SCSI
controller (only one for the whole VM), we're using two raw mapped LUNs
to each VM both OCFS2, we're not using redundant SAN uplinks (so there's
no managed paths), Physical mappings (not virtual). We had some
problems when we first started before we figured out we needed to keep
the VMs on different physical ESX nodes since multiple VMs on the same
host didn't play well with raw mapped physical LUNs (which seems obvious
in retrospect).
In this set up we didn't have to adjust the SCSI host number (as Kent
mentioned). We've run heartbeats on both a private network (second
virtual NIC, dedicated virtual switch in ESX) and the primary network
interface; both have worked fine.
--mark
Mark Sedlock
Network and System Services
Rowan University
-Original Message-
From: [EMAIL PROTECTED]
[mailto:[EMAIL PROTECTED] On Behalf Of Haydn Cahir
Sent: Monday, July 28, 2008 11:08 PM
To: ocfs2-users@oss.oracle.com
Subject: Re: [Ocfs2-users] OCFS2 and VMware ESX
Hi Mark,
Thanks for your reply. How did you configure your RDM mappings? We have
tried a few combinations already. We have three nodes and are trying to
use a single OCFS2 volume. We are encountering a range of errors like
VM's not starting when another node is already started (goes back to the
RDM configurations we think), two of the nodes are able to edit files in
the OCFS2 volumes but the third doesn't see any changes made by the
other nodes and the OCFS2 volume switching to read-only due to errors on
the volume.
We have tried running just two nodes and still get the problem where the
volume will switch over to read-only. I will look into the time
differences on the server, we normally have to make changes in the grub
config and NTP settings to keep the time in synch.
FYI the version of OCFS2 on SLES 10 SP2 is completely different:
HIT-TCN1:~ # rpm -qa | grep -i ocfs
ocfs2-tools-1.4.0-0.3
ocfs2console-1.4.0-0.3
I can't even find reference to this version on the Oracle web site.
Cheers
Sedlock, Mark A. [EMAIL PROTECTED] 07/29/08 12:42 PM
We run a similar set up, SLES 10 SP1, we were ESX 3.0.x and are now 3.5.
We're running the version of ocfs2 that shipped with SLES 10 SP1.
4 nodes accessing raw mapped LUNs via ESX from an HP SAN on HP Blade
Servers. Qlogic HBAs, standard NICs; nothing special.
The biggest hurdle we ran into was time synch on the individual hosts
(VMWare ESX + some variants of Linux have an interesting clock tick
relationship which I still don't understand) that was causing some ugly
fencing.
It's been running well for about 8 months. Overall we're pretty happy
with it thus far. That said, we don't let ESX VMotion the cluster nodes
via DRS, but that's more because we haven't tested it. The cluster is
used for Apache web hosting.
web7:~:%1003#rpm -qa | grep -i ocfs
ocfs2-tools-1.2.3-0.7
ocfs2console-1.2.3-0.7
ocfs2-tools-devel-1.2.3-0.7
web7:~:%1004#uname -a
Linux web7 2.6.16.53-0.16-smp #1 SMP Tue Oct 2 16:57:49 UTC 2007 i686
athlon i386 GNU/Linux
web7:~:%1005#cat /etc/SuSE-release
SUSE Linux Enterprise Server 10 (i586)
VERSION = 10
PATCHLEVEL = 1
web7:~:%1006#
--mark
Mark Sedlock
Network and System Services
Rowan University
___
Ocfs2-users mailing list
Ocfs2-users@oss.oracle.com
http://oss.oracle.com/mailman/listinfo/ocfs2-users
___
Ocfs2-users mailing list
Ocfs2-users@oss.oracle.com
http://oss.oracle.com/mailman/listinfo/ocfs2-users
