[STATSGRASS] multi session slow down

[Roger Bivand]

On Fri, 17 Mar 2006, Joel Pitt wrote:

> Without knowing much about the R interface I would guess that it may be slow 
> due to all machines trying to access data off of the same disk. Have you got 
> any way to measure disk reads and the network bandwidth to the machine 
> hosting the disk?

I agree - I think the original process was also disk-bound, that is 500 
per hour looks very much like 30 times 20 per hour? This would imply that 
10 machines would do 50-60 an hour each. So spreading the compute load 
doesn't help, if this is the case.

Roger

> 
> -Joel
> 
> On Friday 17 March 2006 2:39 pm, rsadler at cyllene.uwa.edu.au wrote:
> > Dear List,
> >
> > I have implemented Monte Carlo inference for a random closed set model
> > that mimics the different "phases" of vegetation patterning to be
> > found in images of a semi-arid grassland in north west Australia. The
> > procedure is implemented using the multiple sessions capability of
> > grass60 on a computing cluster with shared disk space. The problem is
> > that when running a single machine alone I can generate 500
> > simulations an hour. However when I run all 30 machines concurrently
> > simulation rate drops dramatically to 20 sessions an hour for a single
> > machine (all machines are the same).
> >
> > I am first contacting the statsgrass list because the procedure uses
> > the grass/R interface for a number of separate tasks. What I don't
> > know is whether the slow down is a result of the grass/R interface or
> > whether the slow done occurs on the grass side of things where there
> > is some shared file that all sessions use (like .grass.bashrc but not
> > that). The program is run as an R batch file using vanilla and slave,
> > with all output is being written to separate text files. All sessions
> > use different locations and therefore different mapsets.
> >
> > Please advise
> >
> > Regards
> > Rohan Sadler
> >
> > PhD Student
> > School of Plant Biology
> > School of Mathematics and Statistics
> > Bushfire Cooperative Research Centre
> > The University of Western Australia
> >
> > _______________________________________________
> > statsgrass mailing list
> > statsgrass at grass.itc.it
> > http://grass.itc.it/mailman/listinfo/statsgrass
> 
> 

-- 
Roger Bivand
Economic Geography Section, Department of Economics, Norwegian School of
Economics and Business Administration, Helleveien 30, N-5045 Bergen,
Norway. voice: +47 55 95 93 55; fax +47 55 95 95 43
e-mail: Roger.Bivand at nhh.no