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? 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? Thank you. Regards, Raghava.
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
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@osl.iu.edu>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@lists.boost.org http://lists.boost.org/mailman/listinfo.cgi/boost-users
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@osl.iu.edu> 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@lists.boost.org http://lists.boost.org/mailman/listinfo.cgi/boost-users
_______________________________________________ Boost-users mailing list Boost-users@lists.boost.org http://lists.boost.org/mailman/listinfo.cgi/boost-users
Hello Nick, I am new to PBGL. I was looking for a parallel graph processing framework, came across PBGL. When would the new version with AM++ be released? I cannot specifically say what classes/functions I would be using since I am new but I want a directed graph which I can change i.e. add edges. I would also need to associate properties to edges & vertices. This graph would be spread across a cluster. Thank you. Regards, Raghava. On Thu, Sep 16, 2010 at 5:49 PM, Nick Edmonds <ngedmond@cs.indiana.edu>wrote:
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@osl.iu.edu>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@lists.boost.org http://lists.boost.org/mailman/listinfo.cgi/boost-users
_______________________________________________ Boost-users mailing list Boost-users@lists.boost.org http://lists.boost.org/mailman/listinfo.cgi/boost-users
_______________________________________________ Boost-users mailing list Boost-users@lists.boost.org http://lists.boost.org/mailman/listinfo.cgi/boost-users
participants (3)
-
Jeremiah Willcock -
Nick Edmonds -
Raghava Mutharaju