From: Marc L. Smith (mlsmith_at_colby_dot_edu)
Date: Mon Aug 15 2005 - 14:45:56 PDT
Hi Jason, On Aug 15, 2005, at 4:04 PM, jcduell_at_lbl_dot_gov wrote: > You could actually have used just > > SSH_SERVERS="n1 n2 n3 n4" > > I.e., the list is per *process* in the job, not per-UPC thread. > There's > no difference for non-pthreaded jobs, but there is for pthreaded ones. > > I need to document this. Thanks for noticing! When I set SSH_SERVERS as you recommend, I get a runtime error: ~~~ mlsmith@n1:~/upc> upcc -T 8 helloworld2.upc mlsmith@n1:~/upc> upcrun helloworld2 Not enough machines in environment variable SSH_SERVERS to satisfy request for (8). Only (4) machines available: n1 n2 n3 n4 *** AMUDP ERROR: Error spawning SPMD worker threads. Exiting... mlsmith@n1:~/upc> ~~~ > Finally, you can use $SSH_SERVERS, but it's deprecated. $UPC_NODES > (or > $UPC_NODEFILE) is now preferred. When I use UPC_NODES instead of SSH_SERVERS, I encounter the same problem. I've tried to remember, but can't be certain. I don't think this is a new problem with the beta. I think this is what occurred before, and the reason I added two copies of n1-n4 in my nodes list. I went poking around in upcrun.conf -- I may have missed some configuration steps when I installed. Should I have specified: default_cpus_per_node = 2 (currently this is blank) and also: udp_spawn_nodes = n1 n2 n3 n4 (currently, this is blank) Thanks, -Marc