I want to use SortSam of Picard to sort individual WGS data, whose size range is 40~100G. And I use qsub command line to submit my script. Here I am not sure how much memory should I distribute to every job (per individual) in the parameter -l s_vmem. I test several times to check memory usage in one node of our cluster, the memory usage was between 17.5~35G.

PS：Because I have lots of WGS files, it can save my huge time if I can efficiently use the whole memory of the cluster. Of course, the prerequisite is to make sure all of the jobs can be completely done finally.

Thanks for any advice.