03-07-2011 12:29 PM
Lately - I have been questioning about such a layout. Since there will undoubtedly tablespaces that are more hot and active than the others -- would there hyperactivity somewhat curtail performance on the other tablesapces considering under ASM -- everything is stiped accross the same set of disks?
Would it have been better to have seprate ASM Diskgroups -- say for redos and indexes, temp table spaces and high I/O requirement data (on Flash or faster storage) and all the rest on normal wide striped diskgroups?
Our typical DBases are provisioned with anywhere between 80 to 160 50 to 100 GB disks and everytime space is needed, that adds to the "width" of the ASM stripe and induces +ASM to balance out the database data accross.
In a RAW or Cooked storage environment - an Admin/DBA is given leeway in locating and storing tablespaces on appropirate disks but on a single diskgroup ASM stored DB -- there's not much choice.
03-08-2011 02:41 PM
Best way, cooked or raw is to group by database instance or supported application.
Having the containers be meaningful makes it easier to deal with issues.
A big disk pot becomes the 800 ton gorilla that everybody is afraid to try and move.
Owner of ISN Corporation
03-09-2011 05:57 AM
But I do now see most SAP implementations and best practices even recommends up to 5 Di+ASM diskgroups...
03-11-2011 12:27 PM
Also some storage systems will have seperate IO queues per exported LUN. So your target box may have better performance when combining several smaller LUNs via volume manager of the operating system (maybe even with striping!?).
03-11-2011 02:22 PM
Lumping all kinds of Oracle data in one superwide stripe (handled by ASM) I don;t think is a good idea - specially for VLDBs.
I am recommending at least 3 ASM Diskgroups. One for Mainstream Datafiles on normal SAN RAID1 LUNs. Another for super hot and I/O intensive ASM Diskgroup for redo logs, temp table spaces and sort areas - on faster disks or even SSDs and a 3rd for IDX, Conrol files, etc.
03-16-2011 05:16 AM
03-16-2011 05:31 AM
Lotsa luns (LDEVs) here too. 50 to 100 GB each (Open-Vs and XP/Hitachi too obviously) and my current DBs range in size from 4 to 12 TB.
03-16-2011 05:56 AM
03-16-2011 09:53 AM
ASM stripes great I believe but only up to a certain "width". ANd I bleieve lumping all hot, warm and superhot datafiles in one ASM Diskgroup is a bad thing.
03-21-2011 07:05 AM
04-02-2011 09:48 AM
The Oracle philosophy is that SAME will remove all your Hot spots issues.
And also that 2 diskgoups one for the FRA and one for the rest is simple and enough to manage your database.
see slide #13
Of course if you have different type of storage (ie SSD) you may introduce extra Diskgroups.
This seems quite simplistic and may not be the right answer for all configuration.