Re: Exadata high capacity vs high prformance drives
From: Tanel Poder <tanel_at_tanelpoder.com>
Date: Sat, 23 Feb 2013 13:55:31 +0200
Message-ID: <CAMHX9JLKWgRpiSWapHM7W4vxhUyQcKaERBv=cGCK0Ve7f2RWog_at_mail.gmail.com>
I tend to think of high-capacity disks first - and go with high-performance disks by exception. Things have changed a lot since V1 and early V2 days. The flash cache can do so much for you - your disks will be less busy thanks to some reads being satisfied from flash cache. The write-cache helps when you have bursts of random writes (large updates/merges or checkpoints) and temp IO. The flash cache can accumulate these writes and de-stage these to (slower) disks later. So it's harder to justify the need for fastest disks that have 5x smaller capacity.
Date: Sat, 23 Feb 2013 13:55:31 +0200
Message-ID: <CAMHX9JLKWgRpiSWapHM7W4vxhUyQcKaERBv=cGCK0Ve7f2RWog_at_mail.gmail.com>
I tend to think of high-capacity disks first - and go with high-performance disks by exception. Things have changed a lot since V1 and early V2 days. The flash cache can do so much for you - your disks will be less busy thanks to some reads being satisfied from flash cache. The write-cache helps when you have bursts of random writes (large updates/merges or checkpoints) and temp IO. The flash cache can accumulate these writes and de-stage these to (slower) disks later. So it's harder to justify the need for fastest disks that have 5x smaller capacity.
The exceptional circumstances are where your (random write) IO requirements are so high - that you steadily generate more writes than the backend disks can handle, so sooner or later the write buffers would be full and you'd be throttled by the backend disks speed. So the 15k RPM "enterprise quality" disks with their faster seek time and lower rotational latency give you the extra IOPS for random IO.
-- *Tanel Poder* Enkitec (The Exadata Experts) Training <http://blog.tanelpoder.com/seminar/> | Troubleshooting<http://blog.tanelpoder.com/> | Exadata<http://www.amazon.com/Expert-Oracle-Exadata-Apress/dp/1430233923> | Voicee App <http://voic.ee/> On Sat, Feb 23, 2013 at 12:23 AM, Carol Dacko <dackoc_at_gmail.com> wrote:Received on Sat Feb 23 2013 - 12:55:31 CET
> Chris,
> I'll still say, it depends.
> MAA recommends to have high redundancy configured for the disks so that you
> can still have two copies of the data if you take a cell down for patching.
> Therefore, if you configure your disks for high redundancy, take the total
> of storage capacity and divide by 3. That is approximately how much storage
> you have available. So if you did high redundancy and the performance
> drives, your capacity is really reduced.
>
> Here at the University of Michigan we decided to go with high capacity
> drives. Since we are consolidating our environments and have multi-terabyte
> databases, we chose capacity over performance. And we configured for normal
> redundancy.
>
> Since we have not launched anything into production yet, I cannot tell you
> if it made a difference. But I have compared programs running in current
> production and on the Exadata machine and there was a marked difference in
> IO performance. Just confirmed how slow current IO architecture is and how
> fast striping and mirroring can be. We are currently RAID-5 plus
> metro-mirror copying as a DR solution.
>
> For what it is worth!
>
> Carol Dacko
> University of Michigan
>
>
-- http://www.freelists.org/webpage/oracle-l