SC-600 RAID-5 vs RAID 0,1

[Marcos Seoane]

I have a client asking me for the posible perfomance increase in case of change in a SC-600 array of a set of disk. Is there any kind of benchmark or comparison anywhere?

[Eric Sorensen_1]

Hello,

Are you asking about the fc60? If so, you can find specifications here:

http://www.storage.hp.com/disk_arrays/midrange/fc60/index.html

and all of HP's disks and arrays are found here:

http://www.storage.hp.com/disk_arrays/index.html

[Marcos Seoane]

Yes, you are right. Is the FC-60 what I was talking about. The fact is that they have installed Oracle Aplicattions over a FC 60 on RAID-5. Oracle has the general recomendation of using RAID 0/1 when installing Oracle so thats why they are thinking about migrating from RAID-5 to RAID 0,1. And that's the reason why I would like to know if there is any benchmark or comparison anywhere.

[Christopher Caldwell · Accepted solution]

What you're asking for: the performance diff b/w RAID-5 and RAID 0,1 for the FC-60. Can't help you there - I'd imagine the performance you net would be implementation dependent. So the answer is, it depends.

I can give you some real world experiences:

If Oracle had their way, everything would be installed on 2G striped Jamaica's. Unfortunately, the folks who came up with that haven't run many high availability shops.

We've got FC60's and Auto RAIDs. We keep *.dbf on the RAIDs w/o much problem. We keep the most commonly used data pinned in memory - we're not transactional, we're upload/and view, so this works really well for us.

Here's the perf numbers from the user docs:
RAID 0 ? Simultaneous access to multiple disks increases I/O performance. In
general, the greater the number of mirrored pairs, the greater the increase
in performance.
RAID 1 ? A RAID 1 mirrored pair requires one I/O operation for a read and two I/O
operations for a write, one to each disk in the pair.
? The disks in a RAID 1 mirrored pair are locked in synchronization, but the
disk array can read data from the module whose read/write heads are the
closest.
? RAID 1 read performance can be twice that of an individual disk. Write
performance can be the same as that of an individual disk.
RAID 0/1 ? Simultaneous access to multiple mirrored pairs increases I/O performance.
In general, the greater the number of mirrored pairs, the greater the
increase in performance.
RAID 5 ? Provides high read throughput for small block-size requests (2 KB to 8 KB).
? Write performance is limited by the need to perform four I/O operations per
write request.
? Because some I/O operations occur simultaneously, performance depends
on the number of disks in the volume group. Additional disks may improve
performance.
? The I/O performance of RAID 5 benefits significantly from write caching.

As you'll note, the performance of RAID 5 and the performance of RAID 1/0 are contingent on the number of disks in the LUN, etc. The performance will also be contingent on the READ/WRITE behavior of your application.

So from your client's perspective, just because the use RAID 0/1 doesn't necessarily mean that they'll see performance improvement, or if they do see performance improvement that the performance improvement is cost justified.

You can say that given a system that's heavy on the WRITE side of the READ/WRITE equation, a system in RAID 0/1 should outperform the same system using a similar amount of disk space in RAID 5.

Here's the pointer to the doc that gives additional detail:
http://www.hp.com/cposupport/manual_set/lpg28493.pdf

Start reading at page 44.