Re: Data and indexes on different volumes

lower cost SSD (usually based on NAND flash technology) is usually much slower on writes than on reads, and in fact many flash-based SSD products can be (much) slower on writes than typical magnetic disks. if you are considering SSD, choose carefully, and be prepared to spend some money.

as mentioned in another follow-up on this thread, it seems odd to me that you would use SSD to address performance issues with raid-5, particularly if you are bottlenecking on writes. you may get much better bang for your buck by going to raid-1, or (almost) equivalently, using ASM redundancy. raid-5 tends to perform poorly with write intensive workloads. even when supported by NVRAM buffers, if you sustain high write volumes long enough, performance will suffer.

On Tuesday, March 22, 2011, Surachart Opun <surachart_at_gmail.com> wrote:
> SSD is fast on Read reads/writes. So It's great to use for Indexes.You are using Solaris + 11gR2. You can use Oracle Database Smart FlashCache feature, If your system is OLTP (that's useful).
>
> If you use SSD for ASM DiskGroup. It's useful, if you use it for Random Reads/Writes
>
> check http://www.freelists.org/post/oracle-l/How-do-you-use-FlashDisk
>
>
> Best Regards,Surachart OpunOracle ACE, Contributor (Oracle User Group in Thailand)
> http://surachartopun.com
> skye: surachart_op
>
>
>
> On Tue, Mar 22, 2011 at 6:28 AM, Patty Vonick <patty.vonick_at_returnpath.net> wrote:
>
> Hi all, We are currently running 11.2.0.1 on Oracle Solaris v10… RAID-5 config.  We’re hoping to move data around in the near future for many reasons – and will likely migrate to ASM soon as well.
>  In the past, it was recommended that we separate data and indexes on different storage – for optimal performance.  But, we’ve been hearing that is a thing of the past, and there are no performance gains by doing this – especially when running on RAID-5.
>  Our SA has been hearing the opposite – that if we separate indexes from the data, and put the indexes on faster disk (possibly SSD), we could see up to a 2x gain in performance – since the updating of indexes is an “expensive” operation.
>  Any thoughts would be very much appreciated. Thanks!
>

-- 
Cheers,
-- Mark Brinsmead
   Senior DBA,
   The Pythian Group
   http://www.pythian.com/blogs
--
http://www.freelists.org/webpage/oracle-l