Warning about Running with a Single-Processor on a Batch Machine

This problem always comes up when running for a single point, because you can only use a single-processor, but may come up in other instances when you are running with one processor. This applies to all the different ways of running in single-point mode.


A warning for submitting single-point simulations to the batch que when only using one processor. On many machines this will mean using up at least an entire node, and being charged for all the CPU's on that node even if you aren't using them. For example, on the NCAR machine bluefire, there are 32 processors for each node and the batch scripts are setup to have exclusive use of that node (and hence be charged for all 32 processors). There are similar issues on other machines, below we show you what to do when running on bluefire.

To change this on bluefire -- change the following:

#BSUB -q regular
#BSUB -x

#BSUB -q share
so remove the "#BSUB -x" which gives you the entire node exclusively, and change to the share que. One other machines you may have to do something similar, but the particulars depend on the given machine, hence you will need to consult with the system administrators for the given machine you are running on.

Note: Another similar problem on many machines is that some batch ques have a minimum number of nodes or processors that can be used. On these machine you may have to change the queue (in some way similar to the above for bluefire) and possibly the time-limits of the job, to get it to run in the batch que.

Another way to get around this problem is to run the job interactively using USE_MPISERIAL so that you don't submit the job to the batch que. For single point mode you also may want to consider using a smaller workstation or cluster, rather than a super-computer, because you can't take advantage of the multi-processing power of the super-computer anyway.