[Swift-devel] 2 coaster issues

[Michael Wilde]

Mihael, can you comment on two new coaster questions that have come up this week:

1. worker.pl seems to consume a fair bit of time polling. (We need to verify and quantify, but I think Justin is seeing this on Intrepid).

If thats correct, can the polling be done with a select that waits on both the service socket and on signals of SIGCHLD child process termination events?

2. On SGE machines (or other schedulers) that honor requests for multiple slots in a packed/fill manner) we typically get a varying number of slots assigned per node. For manually started workers for persistent servers, can each worker have its own setting of WorkersPerNode, so that a single worker can run the right number of jobs on each node?

This relates a bit to question 1, as we can solve the problem by using WorkersPerNode=1 and starting Nslots workers per node, but if each is polling thats somewhat undesirable.

- Mike

-- 
Michael Wilde
Computation Institute, University of Chicago
Mathematics and Computer Science Division
Argonne National Laboratory