[GRASSLIST:1712] Re: clustering

GRASS GRASS user
1

Mosix is most effective you have multiple processes, so if you were running
the same program hundreds of times with a different dataset this may be good
(think of flooding scenarios). If you are looking for a speedup of a single
complex function and you have a Multiprocessor box thread the CPU intensive
parts of the code. If you have a complex algorith that has little data
dependency between the elements it may be suitable to split over a cluster
using PVM or MPI.
Its really a horses for courses approach.

--
Ian Willis
Systems Administrator
Division of Entomology CSIRO
GPO Box 1700
Canberra ACT 2601
ph 02 6246 4391
fax 02 6246 4000

-----Original Message-----
From: Lars Bromley [mailto:lbromley@aaas.org]
Sent: Saturday, 7 April 2001 7:55 AM
To: rshepard@appl-ecosys.com; ehillmuth@digitalmeridian.net
Cc: grasslist@baylor.edu
Subject: [GRASSLIST:1711] Re: clustering

MOSIX looks like it might allow me to run GRASS on a cluster without
recompiling GRASS for multiple processors, am I right?

I will indeed be trying this in the next week or so. I will be sure to
report to the list what happens. Luckily, I work in a large bloated office
environment which is fully caught up in perpetual windows upgrades, so older
PCs are one thing I have plenty of. Actual GIS education, heck no, but old
computers, no problem :wink:

Lars

Rich Shepard <rshepard@appl-ecosys.com> 04/06/01 04:44PM >>>

On Fri, 6 Apr 2001, Eric Hillmuth wrote:

I had been following the clustering discussion a few days ago and ran

across

this site today:

http://www.mosix.org/

I wonder how useful this kind of clustering framework would be for GRASS
applications. It seems pretty straightforward to set up.

Eric,

  It certainly looks impressively easy! Now, if I had the spare cash to set
up a bunch of linux boxes in a rack, I could test it. :slight_smile:

  Worth looking at it detail, I suggest.

Thanks,

Rich

Dr. Richard B. Shepard, President

                       Applied Ecosystem Services, Inc. (TM)
            2404 SW 22nd Street | Troutdale, OR 97060-1247 | U.S.A.
+ 1 503-667-4517 (voice) | + 1 503-667-8863 (fax) |
rshepard@appl-ecosys.com

2

I sort of figured it would come down to what you wanted to do with your cluster.
Mosix looks like a pretty sophisticated way of moving processes around as well a
more or less easy set up (as in no Grass recompiling necessary) but it also
looks like a Mosix cluster would have pretty limited uses as Ian describes.

Either way I'm interested to hear how Lars's efforts go.

"Willis, Ian (Ento, Canberra)" wrote:

Mosix is most effective you have multiple processes, so if you were running
the same program hundreds of times with a different dataset this may be good
(think of flooding scenarios). If you are looking for a speedup of a single
complex function and you have a Multiprocessor box thread the CPU intensive
parts of the code. If you have a complex algorith that has little data
dependency between the elements it may be suitable to split over a cluster
using PVM or MPI.
Its really a horses for courses approach.

--
Ian Willis
Systems Administrator
Division of Entomology CSIRO
GPO Box 1700
Canberra ACT 2601
ph 02 6246 4391
fax 02 6246 4000

-----Original Message-----
From: Lars Bromley [mailto:lbromley@aaas.org]
Sent: Saturday, 7 April 2001 7:55 AM
To: rshepard@appl-ecosys.com; ehillmuth@digitalmeridian.net
Cc: grasslist@baylor.edu
Subject: [GRASSLIST:1711] Re: clustering

MOSIX looks like it might allow me to run GRASS on a cluster without
recompiling GRASS for multiple processors, am I right?

I will indeed be trying this in the next week or so. I will be sure to
report to the list what happens. Luckily, I work in a large bloated office
environment which is fully caught up in perpetual windows upgrades, so older
PCs are one thing I have plenty of. Actual GIS education, heck no, but old
computers, no problem :wink:

Lars

>>> Rich Shepard <rshepard@appl-ecosys.com> 04/06/01 04:44PM >>>
On Fri, 6 Apr 2001, Eric Hillmuth wrote:

> I had been following the clustering discussion a few days ago and ran
across
> this site today:
>
> http://www.mosix.org/
>
> I wonder how useful this kind of clustering framework would be for GRASS
> applications. It seems pretty straightforward to set up.

Eric,

  It certainly looks impressively easy! Now, if I had the spare cash to set
up a bunch of linux boxes in a rack, I could test it. :slight_smile:

  Worth looking at it detail, I suggest.

Thanks,

Rich

Dr. Richard B. Shepard, President

                       Applied Ecosystem Services, Inc. (TM)
            2404 SW 22nd Street | Troutdale, OR 97060-1247 | U.S.A.
+ 1 503-667-4517 (voice) | + 1 503-667-8863 (fax) |
rshepard@appl-ecosys.com

--
Eric Hillmuth
Principal
Digital Meridian
http://www.digitalmeridian.net
Phone: 303-809-8972
Fax: 303-443-9427