Re: workaround for communicator id

Steve Huss-Lederman (lederman@super.org)
Fri, 11 Aug 1995 13:01:26 -0400

Well, my mail had the intended effect of getting a discussion started.
Several good points have been brought up, including:

- The proposed scheme has problems with inter-comms because the groups
are symmetric and two nodes have rank 0.

- The assumption that only disjoint comms have a problems with unique
ids is wrong. Ids can get reused, etc.

- The work on dynamic processes will cause problems.

Bill brought up another idea we discussed to use the callback feature
when profilers, etc need this information. It is a harder solution
but if it works I would accept it unless these other problems are
overcome.

I'll be offline for a few days. Please continue this discussion and
I'll try and summarize for the meeting next month.

Steve