Re: [igraph] Running Igraph in parallel

[Walter de Back]

Hi,,

If you're implementing in C/C++, you could use openMP to exploit multithreading (on a multi-core machine) to do various analysis tasks on your huge graph in parallel. 

I've made some simple openMP examples showing how to:

- generate and analyse multiple graphs in parallel
http://walter.deback.net/media/c/igraph_openmp1.c

- do different analysis tasks on the same graph in parallel
http://walter.deback.net/media/c/igraph_openmp2.c

More on openMP:
 http://openmp.org
 https://computing.llnl.gov/tutorials/openMP/
 http://en.wikipedia.org/wiki/OpenMP

Best,
 Walter

On Wed, May 28, 2008 at 11:28 AM, Gabor Csardi <address@hidden> wrote:

Lorenzo,

some points. The implementations in the igraph C library are not
parallel, so for igraph it does matter whether it runs on a 4-processor
machine or on a 1-processor machine, it will run with the same speed.

What you can do, however, is to run multiple igraph instances (or
R instances if you use R) on the same machine, and can make the
analysis faster. In R see the 'snow' package on how to do this easily,
this would be the cheapest option, in terms of the amount of work
required.

Another option is to install a batch scheduler system on your
cluster, like 'Condor' 'Sun Grid Engine', 'LFS', etc. THe scheduler
can then "optimally" distribute the work among the machines.
Again, this won't make a single betweenness calculation faster,
but if you do it for 10 graphs then you can run them in parallel.

Best,
Gabor

On Wed, May 28, 2008 at 11:16:45AM +0200, Lorenzo Isella wrote:
> Dear All,
> My institute seems finally to have some budget to spend on hardware equipment.
> This means I could run more intensive simulations, in whose analysis
> the igraph library plays an important part.
> Now, as a consequence, I may end up with the analysis of non-directed
> graphs with e.g. 10^5  nodes.
> Is this doable with the igraph library? And above all: since I am
> likely to generate many of these graphs, can (and how) I use igraph in
> parallel to take full advantage of a cluster?
> Cheers
>
> Lorenzo
>
>
> _______________________________________________
> igraph-help mailing list
> address@hidden
> http://lists.nongnu.org/mailman/listinfo/igraph-help

--
Csardi Gabor <address@hidden>    UNIL DGM

_______________________________________________
igraph-help mailing list
address@hidden
http://lists.nongnu.org/mailman/listinfo/igraph-help