A scenario for the application designers who use LAM and FLTK.
Presently I am running my lam code in which the following happens:
root node:
run the fltk code(a user-interface,(event driven)
with callbacks to run further MPI code.
MPI_Finalize is called on callback of an Exit button
<fltk::run()> the main event loop of fltk.
other nodes (about 4):
wait for root to send messages
process messages and send the results back to root
keep on waiting
but because they wait again,
and root calls MPI_Finalize,they are still waiting.
Question for LAM members:
Is there a signalling mechanism through which the
root process may be able to notify other nodes to quit their
jobs and finalize ?
Question for FLTK members:
After the fltk event loop has begun, can fltk events be addressed
to a different process, other than the code where the event loop
was initiated ?
Now, so far this was my view of designing my application:
---------------------------
| |
| LAM |
| --------- |
| | FLTK | |
| --------- |
|--------------------------
must I try this
---------------------------
| |
| FLTK |
| --------- |
| | LAM | |
| --------- |
|--------------------------
An interesting far-sight:
any thoughts as to how error handling be addressed in such an application?
Waiting for comments,
Thanks,
Gaurav Saxena
Research Assistant
Kent State University
|
