I've been contracting at my current shop for the last four years and everything
has been moved to vm apart from a few ODA's.
I've got 4 four node 4 cpu 12.1.0.2 vm racs along with 70+ vm single instance
using oracle restart.
The racs are all policy managed which I really enjoy and we make sure that we
have one vm rac node per esx host. ie. you don't want all four nodes of a rac
cluster to be on one esx host unless they get there via a multiple host failure.
Storage people started out lumping us with everything else but overtime they've
had to make some concessions.
Direct path read performance is lousy on the virtual disks for all
applications. Some apps it doesn't matter and it's accepted but for those where
it wasn't we moved to prdms (physical raw device mappings). You lose
flexibility on the vm side with prdms. I think there is some restriction on how
many prdms you can have per esx cluster.
All racs are using prdms and you want more smaller disks than one large disk.
You could have a diskgroup with 10 100Gb disk for a high i/o app and a
diskgroup with one 500Gb disk for less intensive apps.
Now you get the joy of having multiple db/vms being affected by underlying esx
host activities/problems.
Mike
________________________________
From: oracle-l-bounce@xxxxxxxxxxxxx <oracle-l-bounce@xxxxxxxxxxxxx> on behalf
of Sanjay Mishra <dmarc-noreply@xxxxxxxxxxxxx>
Sent: February 18, 2016 11:19 PM
To: Oracle-l Digest Users
Subject: Oracle RAC Storage Questionnaire/Recommendation
Hi Everyone
Can someone share any template or points that should be taken into
consideration for doing planning/Architect of SAN storage w.r.t RAC/ASM
infrastructure where we have multiple databases on the cluster. I am giving
some points and so likewise need help in creating the list of best practices to
plan for new SAN requirement for Oracle infrastructure.
1. Should we have dedicated SAN for Oracle environment and not shared with any
other Application/rdbms or so ? Generally it is not economical for an
enterprise to have multiple SAN and so what are the points that need to be
considered to carve LUN utilizing Spindles/pool details.
2. Should we share the Diskgroup among multiple Database like if we have 10
Database on the CLuster or we should isolate some heavy IO/oltp/Warehouse ?
3. What are best recommendation for the number of Diskgroup? I am not taking
into consideration of Sector 4K disk where we might need one extra DG for some
Spfile/Voting/OCR etc. Looking what is best practices and why? One reason is
that there are some RAC setup where saw one High IO database affect all other
application using other database sharing same DG and so if there are multiple
such kind of Database then what are best practices?
4. What are best practices to get LUN carved for ASM ? Ofcourse we will have
same size but as I am not good in Storage terms/technology and need advice the
best practices to update storage to provide LUNS for different kinds of
Diskgroup like Log/FRA/Database etc to avoinf high IO ?
Any other suggestion/advice for Storage is really appreciated
TIA
Sanjay
