Re: the argument with OFA and RAID

From: Frank van Bortel <f.van.bortel_at_vnl.nl>
Date: Fri, 08 Oct 1999 15:23:38 -0700
Message-ID: <37FE6EEA.DBADA76E@vnl.nl>

Verna Legaspi wrote:

> I'm in the learning/transition process of becoming our group's DBA, as our
> Sr. DBA has left the department. Among my main concerns are how our current
> database files are spread out. We're using a Sun SPARC Solaris 2.6 box, and
> all the database files (control files, redo logs, datafiles, etc) are on a
> "SPARC storage array with RAID 0 implemented." (Quoted from our Unix admin)
> My understanding is that this comprises of about 20 2gb physical disks, but
> referenced as 1 logical disk. Is this an "unwise" configuration? Our Unix
> admin's argument against changing how its setup now is that "RAID 0 is the
> fastest, and Oracle's OFA was conceptualized in the days before 2gb HDs and
> RAID 0s, so spreading files across several disks would be irrelevant." Now
> lets consider that our Unix admin knows only enough about OFA and Oracle to
> understand that Oracle recommends an OFA configuration. He doesn't know the
> reasoning behind it. Unfortunately, I am almost in the same boat as him.
> Except I'm a little more open for discussion and argument.
>
> I've read a few discussions about the use and configuration of RAID, and
> I've read the Oracle 8 DBA Handbook about the importance of spreading your
> files across multiple disks, and the "dream" 22-disk configuration.
> However, I don't think I've found any mention about RAID 0 replacing the
> need for OFA.
>
> Any hints, pointers, URLs, recommendations, suggestions, etc. would be
> greatly appreciated.
>
> TIA.
> Verna Legaspi
> vlegaspi_at_uswest.net

As it is seen as one physical disk, files are bound to be spread over multiple disks.
The only point is, you have no control over it. OFA may even be a bad, or non-implementable
idea on some systems. AIX with LVs would be an example. Anyway, maximum throughput is only archieved when allowing maximum IO per second.
Any disk can only handle x IOs per second. Spreading a datafile (table) over multiple
disks (say 5) increased throughput to 5 x IOs per second. But that may also depend on
the array, buffering, etc.

But you do want some control, either by creating Logical Volumes on that Virtual Disk,
or by creating multiple VDs.

Ask your Unix bloke this question: can you garantee I can recover the databases, by applying
the online redo logs, if a disk crashes? If he cannot, change the setup, decline any responsability,
quit, or have him replaced. You do not want tablespaces and redo logs on the same disk!
--
Met vriendelijke groet/kind regards,

Frank van Bortel
Technical consultant Oracle

Work:                                Home:
----------------------------------   ----------------------------
V&L Informatica BV                   Hunzestraat 4
Palatijn 3, 7521 PN Enschede         7555 WB Hengelo
PoBox 545, 7500 AM Enschede          (31)074-2425046

053-4341500 Received on Fri Oct 08 1999 - 17:23:38 CDT