Subject: Re: [Boost-users] [BGL] Upper limits on graph size
From: Jeremiah Willcock (jewillco_at_[hidden])
Date: 2010-05-20 15:58:41
On Thu, 20 May 2010, Bo Jensen wrote:
> Jeremiah,
> Can you say a bit more about the work you mention on BGL for
> shared-memory ? Has some implementation been done ? I ask because I am
> considering to use BGL for parallel computing on shared-memory
> machines.
This is very much work in progress; part of our experimental work handles
threads in limited ways but this is not in Boost's version of PBGL and is
not ready for production use. You can run PBGL with multiple MPI
processes on the same machine, but this is somewhat inefficient. One
issue with shared-memory parallelism in general for graph algorithms is
that memory (both size and performance) is often a limit, rather than CPU
performance. In that case, using multiple cores that share memory is not
worthwhile. More CPU-intensive algorithms do benefit from more cores,
though.
-- Jeremiah Willcock