Re: NT Oracle database with RAID-5 hardware controller - Is it good?

Having done a lot of real world testing on a variety of platforms, I'd add (and *always test your own system for accurate data*):

iolo_at_my-dejanews.com wrote in message <6pi34l$h59$1_at_nnrp1.dejanews.com>...
>> So, in layman's terms:
>>
>> READING: Fast
>
>Read operations will be faster

Reading is generally as good as striped (RAID0), with RAID5.

>
>> Writing: A little slower.
>

Writing can be *substantially* slower. I'd read reports of people seeing 100-300% slowdown on writes for RAID5 v normal striping. I didn't really believe it - until I went away and did my own testing. Be very careful with RAID5 in write operations, the penalty can be much worse than you think. That's why you test.

>> But don't forget one of the primary purposes of RAID 5: Fault tolerance.
if
>> one of the drives fail, your still up and running. And if you have a hot
>> spare, it will automatically bring it on line in the event of a bad
drive.
>>
>
>If complete reduncy of data is required you'd have to use RAID 1.

But this is not necessarily any better than a RAID5 array with a hot-spare. I've had a couple of RAID5 drives go down on me and the hot-spares have rebuilt just fine.

>The striping of data across several drives has other consequences besides
>balancing I/O. One advantage is that logical system files may be created
>which are larger than the maximum size normally supported by an operating
>system

Assuming the OS supports those greater file sizes... And with an LVM you don't have to stripe just to create larger logical volumes. (BTW, I tend to run raw, which I much prefer via a good LVM.)

>the disadvantage is, however, that it is no longer possible to
>locate a single datafile on a specific physical drive. This may cause the
>loss of some application tuning possibilities. Database recovery can also
be
>more time-consuming. If one physical disk needs recovery - all the disks
>which are part of the logical RAID device must be involved in the recovery.

Yep, which is where the RAID1 bit comes in handy (with RAID0).

>
>Datafiles and archives can be placed on RAID arrays since they are accessed
>randomly.

Most datawarehouses use nothing but sequential reading and writing in large batches. Again, I'd suggest a bit of caution here.

>Redo logs should *NOT* be put on RAID arrays since they are accessed
> sequentially and therefore performance is enhanced having the disk head
near
>the last write location. Mirroring of redo logs, however is strongly
>recommended.

Not on RAID5, I think you mean.

>
>
>Oracle recommend RAID 1 rather than RAID 5.

No they don't. *All* the Oracle White Papers and books (and testing) I've read/done show the pro's and con's of each RAID level set-up. You make your own decision based on your needs and budget. Try convincing a customer with a 10TB datawarehouse that RAID1 might be a good idea for all their drives - quite an expensive, and maybe unecessary, move.

>
>
>Having said all that, there don't seem to be any perceptible performance
>problems with our Oracle servers under RAID 5. (DIGITAL PRIORIS / AIX)
>where each individual disk size is 2.5 Gbyte.

If you don't have much heavy write activity then that can be the case. But then have you ever compared and thoroughly tested your system on alternative set-ups (RAID0, 1 or a mix)? That's the only real way you'd know the difference. Most customers don't do that much upfront benchmarking of various configurations of their kit, in my experience.

MotoX. Received on Tue Jul 28 1998 - 02:19:43 CDT