Hi Adi,
We're running a few 9i RAC clusters. Not a lot in production yet, but
we've been stressing and tuning it for over a year. Typically 4 nodes
per cluster, using a combination of OCFS/iSCSI and NFS. Datafiles go
onto OCFS/iSCSI and archive logs go onto NFS (because OCFS sucks at
small files).
We're using Redhat 2.4.9-e.38enterprise. We've been working directly
with Netapp and Oracle development, pulling them tooth and nail along
the cert process, and I believe it just got completed, so we're probably
the first in production. Contact Oracle for more info.
We even have a bi-coastal set of clusters, with multiple masters running
streams replication; as far as I know we're the first at that too. It's
difficult to implement, I wouldn't recommend it.
Note that we are using the Adaptec 7211C TOEs, not the Cisco drivers.
Oh, and by the way, it works great, and many of our NFS DBA's are
drooling over iSCSI at this point. It's much faster than NFS, and much
cheaper than FCP, and when 10gig comes out, it will be much faster *and*
cheaper than either.
There are some Adaptec tuning parameters I can share if people are using
the 7211C's. It helps to raise the inflight cap. Also, there are some
hacks you can do to force linux to mount multiple luns. By default it
only scans the first lun per target (because of a problem with CD
stackers trying to scan every cd at boot). The Adaptec instructions
tell you how to rebuild the kernel, which, of course, defeats your
Oracle support. There is a better way...
The downside is that you don't get the great features that have been
baked into Netapp's NAS software for so long. Remember that iSCSI is
SAN, and Netapp rocks at NAS, but are new at SAN. Snapshots take up a
lot of space, a full lun size per snap or +1 lun + 20%,depending on who
you talk to. Also, the nice .snapshot directory isn't there. You can
implement that with cron jobs quiescing the db,
deleting/creating/renaming/remounting snapped luns, but it's messy.
Bottom line is that if you're considering a move from NFS, the
performance will blow you away, but the maintenance might be a pain in
the arse.
-Mike Christian
-----Original Message-----
From: owner-toasters(a)mathworks.com [mailto:owner-toasters@mathworks.com]
On Behalf Of Aditya
Sent: Friday, June 11, 2004 10:44 AM
To: toasters(a)mathworks.com
Subject: oracle 9i cluster using iScsi?
Anyone have experience(s) running an Oracle 9i cluster with data
stored via iScsi? on a netapp? with Solaris or RedHat?
We already run iScsi for other things and have no experience (or
interest) in DAFS just for Oracle. We're trying to replace Sun 3310
RAID arrays with network-attached storage...
Thanks,
Adi