Re: [boost] Question regarding parallel boost graph library status

Subject: Re: [boost] Question regarding parallel boost graph library status
From: Jeremiah Willcock (jewillco_at_[hidden])
Date: 2011-12-12 23:57:25

On Mon, 12 Dec 2011, Delalondre Fabien Jonathan wrote:

> Hi,
>
> I would like to use the parallel boost graph library for projects I am 
> working on and was wondering whether I could get an update on its 
> development. I am for example wondering:

> 1) Whether this is still an projet under active development and how many 
> people are working on it.

The version that is under active development is a rewrite that has not yet 
been released. That version will be much more scalable and directly 
include thread support.  We have one person primarily working on the graph 
algorithm implementations, and I spend some time on the infrastructure 
parts of the code.

> 2) What the status of the mpi/threading implementation is.

The new version has direct thread support; the current one in Boost can 
only use threads when they are treated as separate MPI processes.

> 3) What are the machines on which the implementation has been tested 
> (and on how many processors).

We are still working on it, but we have tested the infrastructure for the 
new version at larger scales (up to 2k processors on Blue Gene/P) and 
should have at least a prototype of the graph algorithm code tested at 
scale within the next couple of months.  See 
<URL:http://www.unixer.de/publications/index.php?pub=127> for information 
and some performance results on the infrastructure code; breadth-first 
search is one of the benchmarks run there, but it is not using the 
in-progress BGL implementation.

> I am interested in using it on HPC machines such as Cray and IBM which 
> includes more than 100k processors.

That's good -- large-scale users are always appreciated.  Out of 
curiosity, which applications are you interested in using PBGL for?

-- Jeremiah Willcock