| Optimizing IDL on a Beowulf Cluster [message #42527] |
Sat, 12 February 2005 10:21  |
FSD
Messages: 5
Registered: February 2005
|
Junior Member |
|
|
All;
I have access to IDL on a 48 node, 96 CPU Linux cluster. Is it possible
for IDL to make use of multiple nodes?
If so, is there any documentation available on the subject?
Thanks
Frank
|
|
|
|
|
|
|
|
| Re: Optimizing IDL on a Beowulf Cluster [message #45279 is a reply to message #42527] |
Wed, 24 August 2005 06:52  |
Michael C Schrick
Messages: 3
Registered: July 2001
|
Junior Member |
|
|
Frank,
I don't know if this will help or not, but I have written an
interface to LAM/MPI for PV-Wave which I am currently using, which
is available for download at:
http://schrick.fastem.com/beowave/
I have access to an 8-node 3.6 GHz Athlon Beowulf cluster. The
problem is that the length of your analysis programs will grow
to meet the size of your computer. Once you go down this path,
you'll never be able to go back.
I realize that PV-Wave and IDL are not the same and I don't know
how much time or effort your willing to invest, or even how
compatible this package will be with IDL. Maybe, someone else will
respond with a package which is more compatible with IDL.
I'm also interested in how other people are using these types of
tools in a Beowulf environment.
thanks,
Mike
Schrick@fastem.com
FSD <frank@digennaro.com> wrote:
> All;
> I have access to IDL on a 48 node, 96 CPU Linux cluster. Is it possible
> for IDL to make use of multiple nodes?
> If so, is there any documentation available on the subject?
> Thanks
> Frank
>
>
|
|
|
|
| Re: Optimizing IDL on a Beowulf Cluster [message #45341 is a reply to message #45260] |
Sat, 27 August 2005 04:51 |
George N. White III
Messages: 56
Registered: September 2000
|
Member |
|
|
On Thu, 25 Aug 2005, Gianguido Cianci wrote:
> this post brings to mind a question: what happens with the licenses?
> AFAIK, idl uses one license per screen-motherboard combination (so you
> can use dual-proc machines easily). What happens in the case of a big
> mamma cluster? I am not so interested in running idl code in parallel
> but rather having loads of idl jobs run independently on different
> chips... any clues?
Does each job need full IDL? My understanding is that IDL licensing uses
more units on faster CPU's, so in principle, each job in a cluster should
consume some license units. Can the jobs use a runtime IDL
environment or maybe GDL?
--
George N. White III <aa056@chebucto.ns.ca>
|
|
|
|
| Re: Optimizing IDL on a Beowulf Cluster [message #45346 is a reply to message #45260] |
Fri, 26 August 2005 08:24 |
Michael C Schrick
Messages: 3
Registered: July 2001
|
Junior Member |
|
|
If you can, get your licenses setup to key off of the user name,
the machine (host) name, and the display name. Then setup your
serf nodes to all have the same name (i.e. serfxx). The master
node can still maintain a separate name for each node (i.e. serf01,
serf02, ...). When you fire up the cluster, the user/host/display
names for the serf nodes will all be the same and will only require
one license. Thus, two licenses will be required, one for the
master node and one for the serf nodes.
In effect, this is the same thing as having multiple processors on
the single motherboard, which would be a single user/host/display.
There are now multiple processors on multiple motherboards, with
a single user/host/display name.
regards,
Mike
Gianguido Cianci <gianguido.cianci@gmail.com> wrote:
> Hi all,
>
> this post brings to mind a question: what happens with the licenses?
> AFAIK, idl uses one license per screen-motherboard combination (so you
> can use dual-proc machines easily). What happens in the case of a big
> mamma cluster? I am not so interested in running idl code in parallel
> but rather having loads of idl jobs run independently on different
> chips... any clues?
>
> thanks,
> Gianguido
>
|
|
|
|
comp.lang.idl-pvwave archive
Members
Search
Help
Login
Home
