First let me say thanks to Ole Tange for writing such a useful program. I do have one issue that makes me wonder if I am doing something wrong or if it is a bug in parallel.
If the number of jobs I run is less than the number of job slots (hosts * jobs per host) then parallel seems to do poorly at distributing the jobs. I would hope for an even distribution but I don't get that at all. An example: 7 hosts, 3 jobs per host = 21 job slots 11 jobs I would have hoped that each host got at least 1 job, with 4 getting two. But here's what happened: host 1: 3 jobs host 2: 0 jobs host 3: 2 jobs host 4: 0 jobs host 5: 3 jobs host 6: 3 jobs host 7: 0 jobs If it seems crazy to have 21 job slots and only 11 jobs, it's because I use a fixed config of hosts and jobs slots and then feed it different sets (varying numbers) of jobs. Am I doing something wrong? I'm using GNU parallel 20130422. The job distribution that parallel picks is 100% repeatable. Reproduce with: for x in `seq 1 11`; do echo sleep 30; done | ./parallel -j 3 -Shost1,host2,host3,host4,host5,host6,host7 --progress Thank you. -- Allyn Fratkin [email protected] Hewlett-Packard Company http://www.fratkin.com/
