Oracle FAQ Your Portal to the Oracle Knowledge Grid
HOME | ASK QUESTION | ADD INFO | SEARCH | E-MAIL US
 

Home -> Community -> Usenet -> c.d.o.server -> Re: Oracle and Raid setup

Re: Oracle and Raid setup

From: Noons <wizofoz2k_at_yahoo.com.au>
Date: Mon, 13 Jun 2005 02:02:17 +1000
Message-ID: <42ac5c85$0$14447$5a62ac22@per-qv1-newsreader-01.iinet.net.au>


VC apparently said,on my timestamp of 12/06/2005 10:47 PM:

> How so ? My point is that having multiple access paths via independent
> controllers provides true parallelism both for writes and reads. In a
> simplistic model I described in my other message, for a two disk
> configuration, there is about 40% gain in performance for reads and 40 %
> loss for writes. Obviously, with a mixed load you can have either
> performance loss or gain. For an equal proportion of reads and writes,
> performance should be the same as a for a single disk.

Not at all. You are assuming that the OS is capable of somehow discerning from the disk controller where the disk heads are, where they'll be after the next batch of channel programs comes through, what the access pattern from the database will be, and so on. Nothing is further from the truth. Of course it can be done. It just isn't.

> What's different ? Are you saying that modern OS's are dumber than the '85
> VMS and cannot parallelize read acces to a mirrorred set of disks ? Please
> clarify what exactly you mean..

Modern OS's?? Unix (and Linux) have an IO architecture that predates VMS! Cripes, in HPUX you can't even have async IO in cooked device access! No, they cannot parallelize read access to a mirrored set of disks. If you doubt it, have a look at the source code for Linux yourself, it's available in any distro. There is nothing in there to do such at OS level.

The amount of specialised coding and metadata (directly to/from a specific piece of hardware) needed to do that is completely out of scope of a general purpose OS. VMS could afford such luxuries because the only disk controllers it ever spoke to were in-house DEC ones: fixed, known specs. Try that with the plethora of disk controllers available nowadays for a Linux or Unix box and you'll spend the rest of your life writing specialised device drivers. That is why that sort of detail is much better left to the controller itself - or its nearest intelligent cousin: the SAN. Spend the effort where it is more effective.

-- 
Cheers
Nuno Souto
in sunny Sydney, Australia
wizofoz2k_at_yahoo.com.au.nospam
Received on Sun Jun 12 2005 - 11:02:17 CDT

Original text of this message

HOME | ASK QUESTION | ADD INFO | SEARCH | E-MAIL US