Hi,   I am looking for a way to start a
Java programme on a computing cluster that uses SLURM for job
submission. As far as I can tell I cannot ssh to its computing nodes,
which rules out using SSHLauncher.   After
looking through the mailing list thought about using APGAS's NoLauncher
for that task, i.e. starting a number of processes with srun and making
each of them connect to one master process, using the command like (I
have redacted the ip of master node that I use):   srun
 -n 4 -N 4  java -cp "./lib/*:./build/libs/*" -Dapgas.places=4
-Dapgas.launcher=apgas.impl.N -Dapgas.verbose.launcher=true
-Dapgas.master=10.10.10.10 test.Test.    test.Test
 is a simple hello world programme, see the end of this post for its
code. To my understanding, the srun command should start 4 APGAS
processes, each on its own node, with one node accomodating master
process. In fact the logs (the messages in the form of "[APGAS] New
place starting at 10.10.10.10:5701" show that indeed 4 places are
started.    The problems I encounter:  (1) the 4 processes start and this is my 
output:   Hello from place(3)  Hello from place(0)  Hello from place(1)  Hello 
from place(0)  Hello from place(1)  Hello from place(0)  Hello from place(3)  
Hello from place(1)  Hello from place(3)  Hello from place(2)  Hello from 
place(2)  Hello from place(0)  Hello from place(1)  Hello from place(3)  Hello 
from place(2)  Hello from place(2)   I
 am starting to learn APGAS and getting acquainted with its terminology,
 however shouldn't the programme print the message only 4 times, once
for each place?   (2) The programme fails to stop. After printing 16 lines of 
"Hello from..." messages, it just freezes.   What is therefore a 
recommended way of running Java programme with APGAS library on a cluster with 
SLURM workload manager?   Hello world code I used:   public class Test {    
public static void main(String[] args) {          finish ( () -> {           
       for (final Place place : places()) {                  asyncAt(place, 
()->System.out.println("Hello from " + here()));                  }  
         });    }  }   Thanks for your answers!   Best regards  Łukasz
_______________________________________________
X10-users mailing list
X10-users@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/x10-users

Reply via email to