Re: [Boost-users] questions on PBGL

Subject: Re: [Boost-users] questions on PBGL
From: Nick Edmonds (ngedmond_at_[hidden])
Date: 2010-09-16 17:49:46

Hi Raghava,

I'll likely be doing most of the PBGL rewrite to use AM++.  AM++ will be the underlying communication framework, but AM++ itself uses MPI, GASNet, and possibly in the future DCMF/Blue Waters/Portals/etc. as it's data transport layer.  With regards to the PBGL interfaces, I'm planning on doing as much as I can to avoid breaking backwards compatibility at the algorithm and property map interface level.  There will likely be a number of interface extensions though.  The consistency semantics of the underlying data may have to change a bit to support some of the hybrid parallel things I want to do, but the single-threaded case *should* be mostly the same.  If you let me know what you're planning on using I can provide more detailed explanations (at least what my plans are if not what the final form will be).

Cheers,
Nick 

On Sep 15, 2010, at 2:35 PM, Raghava Mutharaju wrote:

> Hello Jeremiah,
> 
> Thank you for the reply and for the pointer to the paper. I had a look at the results using AM++, when would the new version of PBGL be released? Would it be in next month? Would the underlying communication framework be AM++ instead of MPI? If I go ahead and start using the current version of PBGL, would there be much change in the code? I am guessing there would not be much change, because you are changing the underlying communication mechanism. 
> 
> Regards,
> Raghava.
> 
> On Wed, Sep 15, 2010 at 4:12 AM, Jeremiah Willcock <jewillco_at_[hidden]> wrote:
> On Tue, 14 Sep 2010, Raghava Mutharaju wrote:
> 
> Hello Boost users,
> 
> I am looking at Parallel Boost Graph Library (PBGL) to check whether it fits our project needs. I have a few questions on PBGL.
> 
> 1) How does PBGL know about the size of the cluster -- its IP address, name?
> 
> That comes from the MPI implementation -- PBGL doesn't do its own communication; it uses Boost.MPI and thus MPI to do it.  You would need to set up an MPI implementation (such as Open MPI or MPICH) and give it information about your cluster.
> 
> 
> 2) What is the performance of it -- are there any other numbers apart from ones in the paper http://citeseerx.ist.psu.edu/viewdoc/summary?doi=10.1.1.107.5681?
> 
> I don't think there are any posted online right now.  There will likely be some posted within the next month, though.  Note that we are working on a new version of PBGL (not released yet) that will be significantly faster; <URL:http://doi.acm.org/10.1145/1854273.1854323> has a paper with some of those results (including some runs with the current PBGL).
> 
> -- Jeremiah Willcock
> _______________________________________________
> Boost-users mailing list
> Boost-users_at_[hidden]
> http://lists.boost.org/mailman/listinfo.cgi/boost-users
> 
> _______________________________________________
> Boost-users mailing list
> Boost-users_at_[hidden]
> http://lists.boost.org/mailman/listinfo.cgi/boost-users