again on node poses the problem of the "Avakas jungle";
I have to wait for you whereas there is enough room for my calculation.
This is due to the scheduler's list of node.
Then if a node is defective I can't circumvent troubles.
Whereas if I was able to feed the scheduler with my own list of nodes
(detected as free) I would be able to ignore this node.
This becomes very frustrating... !
This is also due to what I called fake parallelized programs, which
indeed, hide N*singlecore calculation (post)-processed by some scripts...
(I am not sure of that) That kind of "hard-used" put the mess on the cluster !
I try to adapt myself and the codes to the Avakas environment but very often
I am facing problems because I am using real mpi/openmp programs... I think
I am not the only one and we may deserve more attention from the admnistrators....