On my linux cluster, I also had a similar problem: the fully qualified host name was required in the machines file. We eventually found a way to solve that. I cannot quite recall how (and the machine is down for the forseeable future due to circumstances outside my control). However, it may be related to how hosts are listed in the /etc/hosts file. Are the hosts listed by their fully qualified name only? Or are there multiple names for each host, one of which is the short name?
Following /etc/hosts would need the fully qualified name in the MPI machines file:
192.168.1.1 node00.cluster.private
192.168.1.2 node01.cluster.private
.
.
.
The following /etc/hosts should allow you to use the short nickname:
192.168.1.1 node00.cluster.private node00
192.168.1.2 node01.cluster.private node01
.
.
.
My apologies for not remembering exactly how this was solved, but it was a year and a half ago that we solved this and my memory of that minor problem (compared to the major porting problems we were having with a 20+ year old legacy code) is rather hazy.
Don Jones
"Unix is simple, but it takes a genius to understand the simplicity."
-- Dennis Ritchie
>Hi again,
>
>>#mpich nodes
>>node00.cluster.private
>>node01.cluster.private
>>node01.cluster.private
>>node02.cluster.private
>>node02.cluster.private
>>node03.cluster.private
>>node03.cluster.private
<snip>
_______________________________________________
scitech mailing list | email@hidden
Help/Unsubscribe/Archives: http://www.lists.apple.com/mailman/listinfo/scitech
Do not post admin requests to the list. They will be ignored.
