Dear BOOST MPI community members:

It was real fun to write a parallel GA solving TSP with BOOST.MPI. Since one of the challenges in writing parallel programs according to Anwar Ghuloum from IBM (http://blogs.intel.com/intellabs/2007/08/03/what_makes_parallel_programmin/) is using modern programming methods BOOST MPI and you as its community are incredible.

Would you be so kind to give me an advice how to measure scalability, detect bottlenecks and in general profile the above mentioned parallel MPI application? My next goal is to generate Gantt chart, but the raw profiling data are perfect. Do you use any tools to verify the bugless execution of your parallel applications, since parallel programs are full with Heisenbugs? Currently only BOOST_ASSERTS are placed around both sending and receiving vectors of user-defined classes over MPI.

I look forward to your reply.

Faithfully yours,

GARO GARABEDYAN, Computer engineer

++359-89-59-59-703

garabedyan.wordpress.com

Sofia, Bulgaria

P Please, consider the environment before printing this email

- Donate spare PC resources to download and compute tasks from the WorldCommunityGrid.org for projects that benefit humanity, like me.