Karl Czajkowski wrote:
On Apr 12, Michel Drescher loaded a tape reading: ...
10) When/why did HostName change from 0-1 to 0-n? (reverted)
At GGF-13 and who keeps trying to set it back to 1? :-)
Guilty as charged. It must have been in the last session that I missed. I am working on releasing the next version of the spec. and it is inconsistent in the definition of this element (some places * and some ?). It was not clear which way was intended or why, hence the comment.
It is 0-n to allow for resource selection of nodes in a parallel job. Chris Smith has that use case with LSF and I can endorse it as a useful thing for parallel jobs too (whether it is in GRAM today or not!). This goes hand-in-hand with the notion of there being a count for the resource element, meaning multiple resources must be allocated to match the same element. The semantics is each resource MUST have one of the names in the list. In other words, if there are more hostnames listed than resources requested, not all names are utilized, or the match predicate for a host is membership of that host's name in the list.
Thanks for the explanation, Karl. I'll have to think about it tomorrow after I get some sleep. (Everything looks better in sunlight. :-) Andreas