Re: commloops cache digest code works

On Sun, Feb 24, 2002, Joe Cooper wrote:

> > Anyway, we're half an hour into a four hour sprint. So far, so good
> > (hits are fine too, at this point).
>
> 2.5 hours+ at 100. No problems, no serious complaints. Hits drop off
> for a few seconds every once in a while, response times jump a lot (from
> 1.5 secs to ~3 secs--not like minutes or anything) every once in a
> while, but overall very very solid. And it's probably measurably faster
> than even 2.2STABLE5+hno. I'll have to run it on a box for which I have
> some specific benchmark data, but I don't think I've seen a single disk
> 450MHz box (with a bunch of crap running on it) maintain 100reqs/sec for
> over two hours with no hit rate degradation or horrible response times.
> I think you've got a winner, Adrian. CPU usage might be lower also,
> it seems like it is, in general--I'll have to try it with two disks to
> see what happens (I've never been able to push a box hard enough with a
> single disk to saturate the CPU, so I don't really know yet).
>
> I'll keep using it as my local proxy for the time being, watching for
> any smelly bits. I'm off to sleep for a few hours, but when I wake I'll
> fire up another run at a higher rate with two disks for Squid.

Ok, cute. It shouldn't be performing quicker - although its slightly
more efficient in its copies, I didn't think it would be very measurably
faster until a few more things were killed.

Thanks for the testing Joe. And thanks for the testing before Kinkie. :-)

Adrian
Received on Sun Feb 24 2002 - 07:19:11 MST