Originally posted by: steevojb

Hi,

I am just having a discussion about Oracle and HACMP. During the Oracle install, what is considered the best practice ?

1. Install the oracle binaries on local disk in rootvg ?
2. Install the oracle binaries on SAN attached disk and have the area move with the database ?

My previous experience is that I have always installed Oracle on local disk and the data areas on the SAN, therefore During a recovery the oracle installation is encompassed in the mksysb.

Any thoughts ?

TIA

Steve
#AIX-Forum

Originally posted by: steevojb

In addition,

I feel it simplifies an hacmp fallover, if all there is to move are the data areas.

TIA

Steve
#AIX-Forum

Originally posted by: SystemAdmin

I will not install it in rootvg. Normally we used to have a separate VG for Oracle and related applications and another VG for data. Of course only the data VG will be part of resource group. Having Oracle on rootvg will unnecessarily increase its size without any significant benefits.

HTH.
#AIX-Forum

Originally posted by: steevojb

Hi,

If you don't install in rootvg, what is your method of recovery in a DR exercise ? a re-install or backup restore ?

TIA

Steve
#AIX-Forum

Originally posted by: ShAKE

In a DR test: last 4 that successfully completed ,I do the following :

Restore mksysb
recreate VG`s,LV`s,file systems (scripting).
change permissions
restore data using netbackup .
let the DBA`s handle database recovery.

I would not recommend installing oracle binaries in rootvg as during the migration you will be not able to import ,export , savevg etc.. if you want to move database to another systems ,along with other reasons.
#AIX-Forum

Originally posted by: orphy

Typically, you don't want to put any non-OS stuff in rootvg so that the mksysb
you cut regularly to be used for recovery is smaller. This means you will have
a short recovery time.

It depends on how you do your DR, with Oracle in non-rootvg VG, you can look
into using savevg/restvg for your backup/recovery. You can always use your
regular backup software to save the Oracle filesystem(s). Or, you can even
tar it off daily. You have plenty of choices here.
Orphy
#AIX-Forum

Originally posted by: KentPerrier

Our situation is probably very different, but we use EMC san with SRDF between our two datacenters. For DR, we simply split the remote copies (the R2s in EMC-speak) from the primary copy in the primary data center, import the VG, fsck, mount and give it to the DBAs to recover.
#AIX-Forum

Originally posted by: orphy

Once you introduce things like SRDF, the whole discussion is changed! Well,
and the money too...

Now the question is whether you want to go to the DR site, with rootvg mirrored,
when all you got is one hdisk went south. To me, it's (usually) a lot cheaper
(both in dollar and (wo)man-power) to just buy the extra hdisk.
Orphy
#AIX-Forum

Originally posted by: SystemAdmin

HACMP is our intention although we do not have now. What happen if I only have 4 HDs divided into rootvg and applvg, 2 HDs of each VG, my Oracle data already on one applvg, one related application on the same applvg, so I put the Oracle engine on rootg, but not the same hd of OS system files, /root, /opt.. etc. IS that bad?
#AIX-Forum

Originally posted by: orphy

Well, the reason to go HACMP is for high availability and possibly appmon,
right?

Now, if you are having two hdisks in rootvg and, as you said, not putting the
Oracle engine on the same hd of /, /opt, etc, does that means you are not
not mirroring your rootvg? If so, that doesn't sound like such a good idea.
I mean, you are spending all that money to put HA together (with redundent HW
and all), I think you should grab more disks and mirror rootvg, preferrably on
both cluster nodes but you might skip the fallover node to save some money if
you have a pure hot-standby config and you have a budget issue.

To answer your question, it's not necessarily bad to put the Oracle binaries
in rootvg but if you have a choice (enough disks), you should put them in a
non-rootvg disk. If you have enough PPs left in the rootvg disks, you should
consider mirroring it. However, you also should consider if doing so will
cause any performance problem with Oracle because of the location of
$ORACLE_HOME.

You should also rethink the whole setup before converting your boxes to a
cluster. Remember, you will want to eliminate as many SPOFs as possible.
Orphy
#AIX-Forum

Originally posted by: SystemAdmin

Here is teh diagram of my rootvg. I beleive the IBM contractor set up to mirror the hdisk0 in this vg, therefore no mirror of my Oracle on hdisk1
rootvg:
LV NAME TYPE LPs PPs PVs LV STATE MOUNT POINT
hd5 boot 1 2 2 closed/syncd N/A
hd6 paging 24 48 2 open/syncd N/A
hd8 jfs2log 1 2 2 open/syncd N/A
hd4 jfs2 1 2 2 open/syncd /
hd2 jfs2 18 36 2 open/syncd /usr
hd9var jfs2 1 2 2 open/syncd /var
hd3 jfs2 79 158 2 open/syncd /tmp
hd1 jfs2 1 2 2 open/syncd /home
hd10opt jfs2 2 4 2 open/syncd /opt
fwdump jfs2 1 2 2 open/syncd /var/adm/ras/platform
fslv00 jfs 234 234 1 open/syncd /oracleapp
paging00 paging 24 24 1 open/syncd N/A
loglv00 jfslog 1 1 1 open/syncd N/A
fslv05 jfs2 79 79 1 open/syncd /oradump
#AIX-Forum

Originally posted by: orphy

Everything is mirrored except for 4 LVs (those with PVs=1). If you have
enough PPs left (lsvg rootvg & look for "FREE PPs" to be at least 338),
you should use mklvcopy to mirror those LVs. If you need a few more PPs to
be able to mirror, you could possibly look into shrinking /tmp (man chfs).
At the minimum, I suggest mirroring paging00 and be sure to back up /oracleapp
and /oradump frequently. You don't want a bad block in paging00 to bring
down your box!
Orphy
#AIX-Forum

Originally posted by: SystemAdmin

Thanks to point out this. from lsvg rootvg, I do have enough "free PPs". /oracelapp is for oracle engine, /oradump is basic somehow an empty place for oracle dump purpose. However I will mirror the paging00. loglv00 is the tracking of the vg, do I have to also mirror all the loglv## on other vg too?
#AIX-Forum

Originally posted by: orphy

loglv00 is the jfslog for /oracleapp which is the only JFS filesystem you have
in rootvg. The rest are JFS2 filesystems which use hd8. Again, since you
have enough free PPs, you should just mirror the two Oracle filesystems. My
guess is that you shouldn't see any I/O issues with the mirroring but you will
be able to reduce any downtime caused by loosing one of the disks in rootvg!

As for "other vg", it doesn't hurt to mirror the loglv* but it's best to mirror
the other LVs as well though you should consider possible performance penalty
with mirroring since it does add a very small overhead to the extra write. I
supposed you can always mirror and then perform some testing to determine if
the small overhead to I/O is worth the minimized downtime.
Orphy
#AIX-Forum

Originally posted by: orphy

I'm not sure if it really "simplifies" a FO. Giving an extra LV or a VG to HA
is not going to make that much different. To me, having Oracle bin in rootvg or
appvg is a matter of managability issue. If you install the binaries on local
(or non-HA controlled SAN disk), you obviously will have to deal with two (for
a 2-node cluster) Oracle installs. This means you will have to patch TWICE,
upgrade TWICE, etc. The upside to this is you likely will minimize app
downtime since you can patch/upgrade one copy, FO your app, patch/upgrade the
other copy, and optionally FB.

Now, if you only have one binary copy, you've just cut your works by 50% though
you will have to worry about more application downtime.

Both are good and both are bad. Whichever to use depends on your requirements.
e.g. I would likely pick #1 if I'm dealing with a 24x7 environment.
Orphy
#AIX-Forum