Re: ASM for single-instance 11g db server?
Date: Tue, 5 Apr 2011 18:36:51 -0700 (PDT)
Message-ID: <16aef091-ca5a-4511-a534-befb008d6645_at_p13g2000yqh.googlegroups.com>
On Apr 5, 4:21 pm, Mladen Gogala <n..._at_email.here.invalid> wrote:
> On Tue, 05 Apr 2011 22:14:34 +1000, Noons wrote:
> > Agreed 100%. Very much so. In fact, I am in the process of
> > re-allocating all our databases to RAID5 in the new SAN, precisely
> > because given two equal SAN/processor environments I can't prove to
> > myself that RAID5 is inherently slower than RAID10.
>
> Noons, that goes against the traditional lore. Do you have any numbers to
> back such claim up?
>
> --http://mgogala.byethost5.com
Hmmm.. I believe that is exactly what I said :) :) If you are using direct-attached disks (PCI RAID controller) I agree that it *may* be slower just as BAARF proposes - but even modern PCI controllers have at least some cache. However, SAN is a completely different ball game. Write performance will be dictated by 1) HBA/SAN switch(es) bandwidth, 2) available front-end cache and 3) CPU/Memory/SGA configuration.
In the past 9 years using SAN, I can say categorically that it was impossible to measure the difference of writes between RAID10 and RAID5/6 when sufficient cache is available and the SAN fabric is properly configured. And as always, YMMV. I can only speak to my experience. And having been in a shops with more than 3000 database servers with more than 4500 database instances and many TB of storage, I do not think my experience should be easily dismissed. You may disagree, but, again, how many people can say they worked in such an environment?
I would suggest that in the end, with the poorly written queries we all deal with, and after looking at all of these "arguments", 99.99% of ALL of the talk of RAID is made completely moot by the abstraction and complications at the application layer. In this day and age of JAVA, Virtual app servers (Weblogic, Webshere,Tomcat, etc. ), and the N-tiered technologies and even RAC, most of the RAID reads OR writes are sufficiently abstracted to the point that in the end it, does it really at all?
The only time you really need to worry about read/write disk-drive performance is when you can "fix" a 1uS delay and that delay is actually measurable at your app level. When was the last time you were actually able to even measure this kind of delay at the end-user level. Maybe in **some** batch processing, but certainly not at the end-user level. Received on Tue Apr 05 2011 - 20:36:51 CDT