[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: transfer and NFS homes
|
From: |
Ole Tange |
|
Subject: |
Re: transfer and NFS homes |
|
Date: |
Fri, 16 Mar 2012 23:42:28 +0100 |
On Fri, Mar 16, 2012 at 12:33 PM, Thomas Sattler
<sattler@med.uni-frankfurt.de> wrote:
>> It is extremely hard to tell the difference between a power user
>> maxing out a server and a novice doing something that overloads the
>> server. I have several times used GNU Parallel causing a cpu load of >
>> 1000, because it was faster to complete my task that way.
>
> I'm curious about that: What kind of job is working better on an
> overloaded system? To my knowledge reaching 100% load (not to be
> mixed with n jobs in parallel on an n-core system) is the best
> to be done.
"Best" depends on how you measure.
In my case I used -j0 because the total time (writing the command +
executing it) would be minimized that way. If I had optimized for
execution time I could probably have found a smaller number of jobs
that would have executed marginally faster, but it would have taken
time to find this number, and thus the total time spend would have
been longer. Each job did a mix of disk I/O, network I/O and CPU in a
hard to predict manner.
/Ole