xfs
[Top] [All Lists]

re: performance over multiple disks

To: Greg Freemyer <freemyer@xxxxxxxxxxxxxxxxx>
Subject: re: performance over multiple disks
From: Chris Tooley <ctooley@xxxxxxxx>
Date: 01 Nov 2002 09:41:05 -0600
Cc: James Rich <james@xxxxxxxxxxxxx>, XFS mailing list <linux-xfs@xxxxxxxxxxx>
In-reply-to: <20021031223250.DRDX3370.imf23bis.bellsouth.net@TAZ2>
References: <20021031223250.DRDX3370.imf23bis.bellsouth.net@TAZ2>
Sender: linux-xfs-bounce@xxxxxxxxxxx
Chalk it up to stupidity or just wanting to learn, but I don't
understand the prime number rationale.  I'm sure there is a perfectly
legitimate reason, but I certainly don't know what it is.  Could someone
enlighten those of us that are not stripe knowledgeable.

Chris Tooley

On Thu, 2002-10-31 at 16:30, Greg Freemyer wrote:
>  >>  On another mailing list a debate arose about performance over a single
>  >>  disk vs. multiple disks.  It goes something like this:
> 
>  >>  Suppose you have a 6 megabyte file stored on disk.  Would it be read
>  >>  faster if it were stored contiguously on a single disk or spread over
>  >>  multiple (say 4) disks?
> 
>  >>  It seems to me that as you get smaller it is faster for the single disk
>  >>  case (remember that we are assuming the file is stored contiguously - not
>  >>  spread all over the disk).  At some size it seems natural that it would 
> be
>  >>  faster if the file were spread over multiple disks.  Can anyone comment 
> on
>  >>  how XFS would perform?  I don't have the equipment available to test 
> this,
>  >>  but I'm not too concerned with actual benchmark numbers.  Mostly I'm just
>  >>  wondering if I understand the filesystem correctly.
> 
>  >>  James Rich
> 
> James,
> 
> I don't fully understand the logic, but I have a Compaq Storage Performance 
> guide in front of me.
> 
> For high data rate applications such as yours it recommends a stripe width 
> (SW) of 17 sectors.  It says that you want this small to get the spindles 
> working in parallel, but if you go below 17, you start getting excessive 
> overhead.
> 
> i.e. 17 sectors of contiguous data per drive.
> 
> For high request rate, it recommends the below stripe widths (SW):
> 
> highly localized requests: SW = 10x avg. transfer size
> 
> highly non-localized requests: SW = 20x avg. transfer size
> 
> unknown localization: SW = 15x avg. transfer size.
> 
> For all cases, they recommend your SW be a prime number.  So the above just 
> get you in the neighborhood and you select the closest prime number.
> 
> Greg Freemyer
> Internet Engineer
> Deployment and Integration Specialist
> Compaq ASE - Tru64 v4, v5
> Compaq Master ASE - SAN Architect
> The Norcross Group
> www.NorcrossGroup.com
> 
> 


<Prev in Thread] Current Thread [Next in Thread>
  • re: performance over multiple disks, Chris Tooley <=