On 7/20/13 3:33 PM, "George Bosilca" <bosilca@icl.utk.edu> wrote:

- The cost of accessing the endpoints will be a load from the ompi_proc_t to get that global index and then another relative load (using this index and the array of endpoints). So exactly the same number of loads as the dynamic case, but one more compared with the "no dynamic support" case in your proposal.

I agree with everything you said, but this is the crux of the problem.  Because you just introduced a dependent (or chained, or however one wants to say it) load in the critical path of the MTLs, which is already heavier than I'd like.  I'm trying to clean it up, but it keeps getting pushed back.  I definitely don't want to make my job harder.

- In terms of memory this solution provide an approach where there will never be an extra overhead. The ompi_proc_t is not changed, and the extra array of endpoints is only created if the components that share it, are all loaded and enabled.

I agree.  Jeff and I talked about a similar concept, but the dependent load was an idea crusher to me.


  Brian W. Barrett
  Scalable System Software Group
  Sandia National Laboratories