[ccan] New module: bdelta

Sam Watkins sam at nipl.net
Thu Aug 25 14:46:54 EST 2011


On Sun, Aug 21, 2011 at 12:18:07AM +1000, Adam Kennedy wrote:
> When your tests can be run in arbitrary environments under conditions
> not under your control, performance testing can be challenging, to say
> the least.

I guess 'mild' performance testing could be done, like timing 100 or 1000
iterations of something simple, or running a test with an alarm that stops it
after a few seconds.  You might want to count CPU time rather than wallclock
time in case the computer is busy doing other stuff.  Anyway I'm sure you've
given a lot more thought to this than I have!

It might be hard to know whether the test passed or not!  Perhaps could compare
the measured performance to some bogus general performance measurement, like
'bogomips'.

We do some more heavy-duty performance / concurrency testing at my work, with
large databases and datasets, etc.  That sort of performance test would not be
good for CPAN-style automatic testing at install time, it might degrade
performance for the whole machine and for a long time.


Sam


More information about the ccan mailing list