Aucbvax.4090
fa.unix-wizards
utzoo!decvax!ucbvax!unix-wizards
Mon Sep 28 19:54:23 1981
profiling kernel changes
>From day@RAND-UNIX Mon Sep 28 19:37:56 1981
I thought of an interesting way to determine the
effect a kernel change will have on performance:
Write your changes like this:
	if (newway)
	    new code with statistics gathering
	else
	    old code with statistics gathering

and have the clock interrupt handler complement
the value of `newway' every interval of, say, 5
minutes.  Since the system load will vary randomly
with respect to your timing interval, after several
intervals you will have statistics for the old
and new ways under approximately the same load.

--dave