2950 write performance issues

NOLLET Remy (CAMPUS) remy.nollet at decathlon.com
Thu Jun 14 02:37:40 CDT 2007


 
On all my PE2950, I have to change default read policy on my Raid disk.
Please check it with "omreport storage vdisk" and change to Adaptative
Read Ahead policy

Regards

Remy NOLLET


Date: Wed, 13 Jun 2007 09:40:18 -0700
From: Sam Flory <Sam.Flory at codegreennetworks.com>
Subject: Re: 2950 write performance issues
To: "McDougall, Marshall (FSH)" <Marshall.McDougall at gov.mb.ca>
Cc: linux-poweredge at dell.com
Message-ID: <46701DF2.5050506 at codegreennetworks.com>
Content-Type: text/plain; charset=ISO-8859-1

McDougall, Marshall (FSH) wrote:
> 
> I keep looking for something I can identify as the problem here.  I 
> just noticed that memory usage is drastically different in these
machines.
> Outside of adjusting the shared memory for these boxes, they are the 
> same.  Can anyone enlighten me as to why there is such a radical 
> difference in memory usage?  What can I look at to tune?
> 
> Server 1 - bad:
> [root at fsh1166db01x]# free
>              total       used       free     shared    buffers
> cached
> Mem:       8310532    8294836      15696          0       6280
> 8144720
> -/+ buffers/cache:     143836    8166696
> Swap:     16386292        208   16386084
> 
> Server 2 - good:
> [root at fsh1166db02x]# free
>              total       used       free     shared    buffers
> cached
> Mem:       8310532    2952540    5357992          0     125464
> 2679676
> -/+ buffers/cache:     147400    8163132
> Swap:     16386292        208   16386084
> 


  The memory differences are a non issue.  The case of the "good" system
you have simply done less io.  Linux considers free memory as wasted.
So as you read from the disk it throws everything into cache.  If you
look at the "-/+ buffers/cache:" lines both systems have the same amount
of real world used, and free memory.  Cache pages in memory are freed if
they are needed.

PS- If you run "cat /dev/sda >/dev/null" for a few seconds on the "good"
system" free should look about the same as the "bad system".




More information about the Linux-PowerEdge mailing list