Print

Print


Something that we found is that (over a network especially) the display
speed could be
a problem.

We recently installed a dual processor opteron machine, with redhat ent.
4.0.  I had the
machine installed  in our university's "Center for Computational
Research" (CCR) (I figured they
wouldn't even notice maintaining another machine ;) ).  Any rate, the
originally performance
was miserable, and had CPU utilization of only 4-5%, for the SPM
coregistration routine.

We were able to deduce that the culprit was displaying over the
network.  Where I'm located
should be 100 Mbit but often we're only about 10-20Mbit.

We considered commenting out most of the display code in some of the
routines, but I was
in a mood and pushed for  finding a better solution.  The CCR folks
tried running a vncserver on
the SPM machine, I installed tightvnc client on my machine.  The
performance increase for
running the SPM coregistration routine was about 20 fold.

I now connect to the new machine using vnc for all of the programs we
use on this machine.  Everything
runs much much better, and vnc offers a lot of other nice features as well.

Something that you might consider is running one of the bench mark
routines and seeing if any
particular area stands out.  In my case, originally the graphics
benchmark was way out of line
with machines of similar processing power.


dave



Brambati, Simona wrote:

> Dear SPMers,
>
> I'm using a Precision Worstation 3.6GHz and 3.0 GB of RAM (DDR). The
> Hard Drive is 140GB with 10GB remaining. Hyper-Threading is on. The
> system is Microsoft Windows XP.
>
> When I run an estimation, I run into a very noticeable problem,
> specifically on the second part of the analysis. The Processor/CPU
> usage hangs at 50% and does not want to use anymore processor. I've
> run the exact same analysis on a different and less powerful machine,
> and the second stage not only shows quicker processing, but the CPU
> usage works 90%-100%.
>
> What can I do to get my machine to work closer to 90-100% on the
> analysis?
>
> Thanks,
>
> Regards,
>
> Simona Brambati
>
>
>