thanks for your answer ... I am beginning to understand - even though I
am not happy with it :)
Eugene Loh schrieb:
> Jens wrote:
>> Hi Terry,
>> I would like to run a paraview-server all time on our cluster (even
>> though it is not in use 24h) - but this would simply result in some kind
>> of "heating-thread".
>> Even though it has (in theory) no impact on the node performace (which
>> is part of a grid-engine), it would simply result in some kind of
>> "heating-thread" which is not a nice idea.
> Further (and you can test this), it doesn't yield 100% effectively.
> E.g., even if you had other processes ready and eager to consume the
> CPU, it seems (based on my tests) that a waiting/yielding OMPI process
> won't give up the CPU very well. E.g., maybe only 50%.
> An MPI implementation could be more cautious. E.g., it could put a
> process to sleep and poll only occasionally for in-coming messages or
> have some other way of waking up when needed.
> The trade-offs are:
> *) the more you give up the CPU, the longer it takes to wake up
> *) the more control over this trade off you give to the user, the more
> the user needs to understand the tradeoffs
> *) the more control you want to allow, the more complex the MPI
> Regarding the last one, I think OMPI only allows aggressive or else
> yield. No sleep.
> users mailing list