[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [Swarm-Modelling] Question about Distributed Simulations
From: |
docgca |
Subject: |
Re: [Swarm-Modelling] Question about Distributed Simulations |
Date: |
Tue, 22 Aug 2006 07:50:06 -0400 |
Hi Marcus,
Thanks for the quick reply. So if you don't mind let me see if I can
describe this in lay terms:
Each node is then responsible for some portion of the overall model (as
far as the computations associated with that portion), then they output
at the end of each "step" to the shared HDF5 database, the data then
being re-inputed into each node at the beginning of each step.
Therefore the communication between each node's portion of the model is
via the database and not from node to node. Therefore optimizing the
structure of the model in the cluster depend on balancing the rate of
information reading/writing into the database, the computational
intensity at each node, and the number of nodes. Furthermore, any
computational process that can read/write into the HDF database can be
parallelized in this way. Is this the general gist?
Gary
-----Original Message-----
From: address@hidden
To: address@hidden
Sent: Tue, 22 Aug 2006 5:00 AM
Subject: Re: [Swarm-Modelling] Question about Distributed Simulations
Hi Gary,
> I am looking into the feasibility of running a Swarm simulation on >
IBM's Blue Gene L, which has as its base rack a 1000 node cluster >
(2000 PPC 5 processors). The question is whether an individual run >
could be done across the entire cluster, with groups of agents >
distributed among all the nodes and communication across the nodes
One way you could think about doing this is to have all of the nodes
of the cluster read and write via a fast shared parallel database such
as HDF5. Swarm has the basic support for reading and writing objects
via HDF5. To have a 2d space, for example, would mean making the
`moves' be updates via a key (an x,y coordinate). There's isn't a
builtin class to do it for you, but it wouldn't be hard to write.
http://hdf.ncsa.uiuc.edu/hdf5-quest.html#phdf5
As HDF5 was developed with DOE/ASCI funding, it is probably already
installed on the Livermore machine..
In parallel programming, this approach is sometimes called a "Tuple
space". http://en.wikipedia.org/wiki/Tuple_space
http://www-unix.mcs.anl.gov/dbpp/text/node44.html
Marcus
_______________________________________________
Modelling mailing list
address@hidden
http://www.swarm.org/mailman/listinfo/modelling
________________________________________________________________________
Check out AOL.com today. Breaking news, video search, pictures, email
and IM. All on demand. Always Free.