Thanks Bill.
I think the simplest thing to do is a pbs_sched option, and it also fulfills the multi-sched use case that @scc highlighted.
A good default, I know everyone has been talking about how it’s a bad idea to launch as many threads as the number of cores/hyperthreads, but TBH, in my testing with a 32 core machine, I’ve not seen the scheduler take more than 10% of the computing power, so I actually think that that’s a good default, otherwise we might not really see much of a performance boost. In fact, I think that the scheduler shouldn’t use multi-threading until we have at least 4-8 cores to work with. I don’t know how many other services get run on the host node, but if PBS gets the majority of the resources then why not use as much of them as we can?