gettimeofday is too coarse
	can't measure low latency
	can't measure high bandwidth
	lat_connect
	small reads
hardware timer hooks
	cold cache
	Report cold cache #s?
statistical methods
	inconsistent in treatment
		claim is: min of a few iterations
		Facts
			16/19 are fixed sized loops
			1/19 is start/stop/sleep
			1/19 is max
			1/19 is min of multiple repetitions
warm/cold cache results
	run short to warm cache
	then time
	Not an issue for us because we run longer
	Autosizing loops handle this
context switching
	claim ours is bad
	working set screws the results
	they took out working set
	our numbers were stable w/o working set as well
