Test impact of memory size on performance

[standage]

trafficstars

I've always been thinking in terms of "more memory is always better because lower FPR". This is true, but there's evidence, at least for some computations, that runtime scales with memory consumption. So there's an optimization to be made here: what is the smallest amount of memory I can request that won't lead to big problems with FPR?

If this observation holds and if @ctb and I aren't completely of our rockers, we should actually get a speedup when we parallelize (as in #15). If we run all the parallel processes at once, then the total memory consumption will still be the same, but we would have the option to split it up.