Amanda-Users

Re: How to omit holding disk?

2004-02-04 01:18:48
Subject: Re: How to omit holding disk?
From: Jon LaBadie <jon AT jgcomp DOT com>
To: amanda-users AT amanda DOT org
Date: Wed, 4 Feb 2004 01:14:12 -0500
On Wed, Feb 04, 2004 at 12:10:17AM +0100, Josef Wolf wrote:
> On Mon, Feb 02, 2004 at 05:57:16PM -0800, Jay Lessert wrote:
> > On Mon, Feb 02, 2004 at 11:39:24PM +0100, Josef Wolf wrote:
> 
> This top output is on idle system:
> 
>   top - 23:19:51 up 5 days, 12:08, 23 users,  load average: 0.00, 0.08, 0.17
>   Tasks: 141 total,   1 running, 137 sleeping,   0 stopped,   3 zombie
>   Cpu(s):   1.9% user,   2.2% system,   0.0% nice,  95.9% idle
>   Mem:    514224k total,   475768k used,    38456k free,    50224k buffers
>   Swap:        0k total,        0k used,        0k free,   182920k cached
>     PID USER      PR  NI  VIRT  RES  SHR S %CPU %MEM    TIME+  COMMAND
>    4676 root      15   0 61376  19m  712 S  1.3  4.0   0:59.98 X
> 
> 
> This one is when copying:
> 
>   top - 23:22:41 up 5 days, 12:11, 24 users,  load average: 2.86, 1.13, 0.54
>   Tasks: 143 total,   3 running, 137 sleeping,   0 stopped,   3 zombie
>   Cpu(s):   1.8% user,  22.5% system,   0.0% nice,  75.6% idle
>   Mem:    514224k total,   509680k used,     4544k free,     8456k buffers
>   Swap:        0k total,        0k used,        0k free,   275012k cached
>     PID USER      PR  NI  VIRT  RES  SHR S %CPU %MEM    TIME+  COMMAND
>   27744 root      16   0    88   88   24 R 12.5  0.0   0:14.56 cp
> 
> As you can see, "cached" goes up (the contents of the huge file get cached)
> while "free" and "buffers" go down. Everything from "buffers" gets flushed.
> The 182MB from "cached" get flushed and replaced with the contents from the
> copied file as well.
> 

My eye focused on something else.  On an "idle" system you had 38MB free of
512MB total.  During copying that dropped to just 4MB free.  My Solaris system
also has 512MB total, but 120MB free when "idle".  Doing a copy such as you
report only dropped it by about 25MB.  I saw little degradation in interactive
performance.  This may be the case where the old saw "add more memory" may be
a valid recommendation.

Does your mount option have a "delayed atime update" option?  That can have a 
big
impact on disk performance on a Solaris system.  Instead of read a block, update
the atime in the inode, read a block, update the atime ...  It delays updating 
the
atime in the inode until file close.  Its impact is most acute on mounts using
the "journaling", aka "logging" option (again on Solaris, YMMV).

-- 
Jon H. LaBadie                  jon AT jgcomp DOT com
 JG Computing
 4455 Province Line Road        (609) 252-0159
 Princeton, NJ  08540-4322      (609) 683-7220 (fax)

<Prev in Thread] Current Thread [Next in Thread>