Creating so many communicators is not a good idea. Why can’t all processes open one large window? — Pavan On May 21, 2014, at 4:10 PM, Ted Sariyski <[email protected]> wrote:

Hi,

I am working on a long-range interaction problem, so that communication is "each-to-each", although very sparse. My current approach is as follow: 1. If a process owns an object A, it opens a shared window to this object. 2. If another process happens to generate data which belongs to A, which is determined on the fly, it 'put' its share to A.

The algorithm works for small number of objects, but I need millions of them. I wonder: 1. Is it possible to increase the number of communicators above 2048? 2. Is it a scalable approach?

Again, although that communication is 'each-to-each', it is extremely sparse. Any advise and comments are highly appreciated.

Thanks in advance, --Ted

_______________________________________________ discuss mailing list [email protected] To manage subscription options or unsubscribe: https://lists.mpich.org/mailman/listinfo/discuss