Path: news.easynews.com!easynews!priapus.visi.com!news-out.visi.com!hermes.visi.com!eusc.inter.net!newsfeed01.sul.t-online.de!newsfeed00.sul.t-online.de!newsmm00.sul.t-online.com!t-online.de!news.t-online.com!not-for-mail
From: Rick Denoire <100.17706@germanynet.de>
Newsgroups: comp.databases.oracle.server
Subject: Need advise about expanding storage
Date: Thu, 06 Jun 2002 23:54:57 +0200
Organization: T-Online
Lines: 41
Message-ID: <bdlvfu8f584vvram0htn6boa2754cu0smc@4ax.com>
Reply-To: 100.17706@germanynet.de
Mime-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Transfer-Encoding: 7bit
X-Trace: news.t-online.com 1023400492 06 9770 g4o1bqYTTTAKO- 020606 21:54:52
X-Complaints-To: abuse@t-online.com
X-Sender: 320058382570-0001@t-dialin.net
X-Newsreader: Forte Agent 1.7/32.534
Xref: easynews comp.databases.oracle.server:149800
X-Received-Date: Thu, 06 Jun 2002 14:52:29 MST (news.easynews.com)

Hello

I am running an Oracle DB (8.1.7, Unix). All Oracle data is contained
in three independent devices. It happens that three devices are to few
to cope with the numerous I/O that Oracle does to different datafiles
and that could be done in parallel.

The question ist: How should be a Raid system configured to be used
with the DB? It is a StoreEdge A1000 (SCSI) with 12 HD that is able to
be setup in different Raid flavours.

I first thought that just making one device out of it by striping data
over all 11 harddisks (1 is for parity) would be fine, since that
would accelerate access a lot. But since Oracle will be trying to
access different files more or less at the same time, having one
device would force serializing access to it. OK, access will be fast,
but can only happen one after the other.

Another possibility would be to make one device (or LUN) out of each
harddisk, so access to each particular harddisk won't be so fast, but
it can be done in parallel, since each harddisk behaves independently.

Perhaps a kind of compromise between these two extreme cases would be
the optimal solution, that is, building several groups of harddisk
which use stripes but behave as independent devices.

Well, having three copies of the controlfile, five redolog groups with
two members each and running in archivelog mode demands a minimum
number of 6 devices, without considering the access requirements for
the datafiles containing the real data.

One hint, however: The datafile in the index tablespace shows a much
larger number of read/write access compared with other datafiles, and
the main datafile containing most objects is mainly being read, but
seldom written to (data mining application).

Need some advise, since I am not that experienced with performance
issues due to the physical topology of the database.

Thanks a lot
Rick