[torqueusers] Node np parameter adjusted automatically since 2.1.x ?

garrick at speculation.org garrick at speculation.org
Thu Jun 15 02:48:39 MDT 2006


On Thu, Jun 15, 2006 at 10:23:27AM +0200, Martin Schafföner alleged:
> Hi all,
> 
> we have a bunch of dual xeon nodes which have hyperthreading enabled (it 
> actually gives some performance improvement especially when memory-bandwidth 
> limited apps run). So there are virtually 4 processors in each node, but we 
> used to limit the number of processes by setting the np=2 parameter in the 
> server_priv/nodes file.
> 
> After upgrading to 2.1.x (actually 2.1.1snap), I saw 4 processes being 
> scheduled on these nodes. Upon investigating, I found that the 
> server_priv/nodes file have automatically been adjusted to np=4 for the 
> respective nodes. That was not what I wanted. Is there any way to prevent 
> this behavior and fix this parameter to what the admin says rather than what 
> the moms say? 

*sigh*  I knew someone wasn't going to like that change.  The idea is to
reduce the amount of initial configuration for new clusters.

There's no server attribute to disable, but I'll add one before 2.1.1
releases.

Just comment out the 'if else' block at lines 1085-1101 in
src/server/node_manager.c.

 
> I know, mom is always right...

That's what I thought too!

The current logic is to only change np if MOM's ncpus is greater than
np.  Maybe a better logic is to only change if (np==1 && ncpus>np)?



More information about the torqueusers mailing list