From: Ben Rockwood (benr@cuddletech.com)
Date: Tue May 27 2003 - 17:16:30 EDT
Hello Managers.
I'm seeking some input from managers who have integrated/upgraded
HA-Oracle under Sun Cluster 3.0. I've gone back and forth on diffrent
ideas of setup and feel I need some input from managers who have
experience and may have already dealt with some of these questions.
1) Installation of the Oracle Binaries. In our current production cluster
the binaries for oracle were installed on the root disks of each node.
The inherent problem I see with this (I didn't set this up) is that the
root disk, imho, should idealing in a cluster node be completely
replacable. Should a root disk suffer catastrophic failure including its
mirror, I don't want to have to replace the oracle installation. One way
of looking at this is that once a base OS is re-installed (if you can't
restore directly from tape to the raw disk) you'll just restore Oracle's
home and binaries from tape anyway. However there is still the question
of putting the IO burden of the binaries on the root disk, although lite
in most operating cases. This still just doesn't seem like an ideal thing
to me. Should I just suck it up and put Oracle on the root disks or put
it on my disk arrays? Arrays seem like the right thing to me.
2) I'm moving from SC2.2 to SC3.0, and sadly from VxVM to SVM. Further
instead of using photons we're using a pair of 3510FC's. The ideal
solution to me seemed to be to export idential luns on each 3510 and then
mirror one lun against the other (array to array) in software via SVM
which is going to be used with SC3.0 anyway. The first LUNs would be
single disk (72G) for Oracles binaries, and then 3 2disk LUNs striped and
mirrored in SVM for data, indexes, rollbacks, etc. This seems like a
pretty solid plan to me, but maybe there is a better way. Even with dual
controllers I'm not comfortable seeing a single 3510 (or purple for that
matter) as being so highly avalible as to trust that I can use a single
stack of them as a shared storage solution in the cluster.
3) Making the oracle binaries global. Is this a particularly good idea?
A centralized installation is nice because it minimizes admin time for the
DBAs and reduces the chances of the installation on each nodes becoming
unsyncronized and causing trouble down the road. Add to this that my
DBA's want to use the Oracle standard convensions (/u01, /u02, etc) and
telling them to use global mount points (/global/orads/u01) isn't going to
make them happy. This could be handled thru symlinks, but that somehow
doesn't seem like a wise idea. My DBAs and manager currently favor using
two seperate installations, largely I think because thats what we've
always done, but I'm not sure if I should try and push (or force them)
into considering global.
Can anyone with experience and opinions on the above thoughts please
share your thoughts with me? I'd greatly appreciate it.
benr.
-- //Ben Rockwood - UNIX Systems Admin //email: benr@cuddletech.com //web: www.cuddletech.com //-> We do what we can, We give what we have, //-> Our doubt is our passion, and our passion is our task, //-> The rest is the madness of Art. //-> -Henry James _______________________________________________ sunmanagers mailing list sunmanagers@sunmanagers.org http://www.sunmanagers.org/mailman/listinfo/sunmanagers
This archive was generated by hypermail 2.1.7 : Wed Apr 09 2008 - 23:26:28 EDT