Thread Subject:

Parallel Computing on Clusters?

Very simple, perhaps obvious(?) question. For basic matrix operations (e.g. C = A .* B), I know that MATLAB does parallel computing automatically e.g. in multicore machines. Is the same true for clusters (e.g. on a server)? Does it depend on the server?

Mostly, I am trying to figure out if my code--which is mostly basic matrix operations with a solver--can be sped up on such a server if I use a parallel approach.

Thanks!

Cory

"Cory " <corydeburd@yahoo.com> writes:

> Very simple, perhaps obvious(?) question. For basic matrix operations
> (e.g. C = A .* B), I know that MATLAB does parallel computing
> automatically e.g. in multicore machines. Is the same true for
> clusters (e.g. on a server)? Does it depend on the server?
>
> Mostly, I am trying to figure out if my code--which is mostly basic
> matrix operations with a solver--can be sped up on such a server if I
> use a parallel approach.

Use of a cluster is explicit and is supported by Parallel Computing
Toolbox in combination with MATLAB Distributed Computing Server.

See here for more:

<http://www.mathworks.com/products/parallel-computing/>

In short, you might be able to speed up elementwise operations using a
cluster. Whether any particular application speeds up using a cluster
depends on the ratio of the time taken to send the work to the cluster
and get the results back versus the time taken to do the
computation. For a single "C = A .* B", sending the data to a cluster
and then retrieving the result will not be efficient because the
transfer time will dominate. If however you have a series of
computations that you want to do, and you can keep the data on the
cluster, then you can get speedup.

Cheers,

Edric.