Re: RAID 5 7+1 with oracle

  • From: "Jonathan Lewis" <jonathan@xxxxxxxxxxxxxxxxxx>
  • To: <oracle-l@xxxxxxxxxxxxx>
  • Date: Fri, 5 Oct 2007 09:44:10 +0100


I'm not convinced your interpretation of the graphs is correct.


Left graph p9:  Total for 4 disks (2d + 2d)

Writes on the 10K rpm start to hit their catastrophe point at 400 IO/s
Writes on the 15K rpm start to hit their catastrophe point at 600 IO/s

Right graph p10: Total of 8 disks (7D + 1P)

Writes on the 10K rpm start to hit their catastrophe point at 400 IO/s
Writes on the 15K rpm start to hit their catastrophe point at 600 IO/s

Exactly the same limit despite having twice the spindles.

If you want to compare write TIMES, you should be looking at
the right graph on p9: 4D + 4D (eight disks) which get to the catastrophe point at 800 and 1200 IO/S


The same applies with read TIMES - surely you should be comparing
the two right-hand graphs.  (I'm a little surprised by the similarity of the
graphs for reads - in a simplistc queueing theory model I would approximate
the RAID-10 as 4 x M/M/2, and the RAID-5 as 8 * M/M/1 with 7/8ths
load - and expected the M/M/2 to outperform the other). But maybe it's not
really valid to call RAID10 M/M/2; maybe hp just have some dirty tricks (or errors) in their algorithms that even things out.


If you want to compare volume of data stored, of course, that's a completely
different matter.



Regards

Jonathan Lewis
http://jonathanlewis.wordpress.com

Author: Cost Based Oracle: Fundamentals
http://www.jlcomp.demon.co.uk/cbo_book/ind_book.html

The Co-operative Oracle Users' FAQ
http://www.jlcomp.demon.co.uk/faq/ind_faq.html

Date: Thu, 4 Oct 2007 17:35:14 -0700
From: "Greg Rahn" <greg@xxxxxxxxxxxxxxxxxx>
Subject: Re: RAID 5 7+1 with oracle

In order to avoid a religious war I'll just offer this:

Look at:
http://h71028.www7.hp.com/ERC/downloads/4AA0-7923ENW.pdf

This document shows that RAID5 (7D+1P : page 10 right graph) is far
better than RAID 1(2D+2D : page 9 left graph) both on read and write
operation.


--
//www.freelists.org/webpage/oracle-l


Other related posts: