Re: RAID

From: Kevin Loney <Kevin.Loney_at_astramerck.com>
Date: 1998/10/07
Message-ID: <JwQS1.93$z07.362923_at_newsread.com>#1/1


Brian Yan wrote in message <36198879.2E53_at_gpu.srv.ualberta.ca>...
>Hi there,
>I am new on DBA stuff. I read DBA Handbook by Kevin Loney. It says that
>some database file should be at the different disks.

Hi, Brian. I hope you've found the book to be helpful.

>My questions are:
>
>1). Does the 'different disks' here mean the different PHYSICAL disks?
>
It means different physical disks.

>However, is this still true for RAID disks?

Ideally, you'd separate across RAID arrays as well. For instance, I have a RAID array for
tables that is separate from a RAID array for the associated indexes. Obviously, the size of
the database and available hardware influences your architecture decisions.
>
>For example, I have five RAID5 hard disks u1~u5. I create DATA
>tablespace in u1, INDEX tablespace in u2. However, when data in DATA are
>stored into hard drive, it will be striped into different drives
>according to RAID. Same thing happens to INDEX data. In this case, won't
>the data in DATA and data in INDEX be mixed up? How can we dedicate the
>whole disk to DATA tablespace under RAID environment?
>

You're using the O/S to stripe the data across multiple disks, so you've distributed the I/O costs across multiple disks. The I/O burden for a tablespace is thus distributed across all of the disks in the RAID set. The I/O cost for an index tablespace on the same RAID array is similarly distributed. So RAID helps reduce the likelihood that you'll max out the IO available for any one disk. There's a potential that there will be a conflict during disk accesses, but since the data is striped at the block level and the parity block is random in RAID5, conflicts don't appear to be much of a problem. I run my largest systems on RAID5 without much conflict. RAID 0+1 is still preferable to RAID5 but may be prohibitively expensive for you.

Hope this helps.

Kevin Loney
http://www.kevinloney.com Received on Wed Oct 07 1998 - 00:00:00 CEST

Original text of this message