Just got asked to take a look at a Smufin process that had been running for 3 weeks without apparently achieving anything. Turns out it was launched with -np 1 but --cpus_per_node=96. Clearly the user expected this to use one *core* on the machine as the master and the other 95 as slaves, but of course in fact the master is always single-threaded and only the slaves (the non-existent ranks 1+) would actually use that setting.
The user hadn't noticed the problem however as Smufin busy-waits with zero free CPUs, rather than exiting promptly when NUM_NODES ends up zero. I suggest you should add a check there.
Chris