[camkego]

Try benchmarking OS read() calls vs. either sequential or random reads using memory mapping, whenever I do this OS read() calls end up being quite a bit faster.