time test

[David Relson]

At 07:42 AM 11/24/02, Greg Louis wrote:

>On 20021123 (Sat) at 1807:40 -0500, David Relson wrote:
>
> > For laughs, I built bogofilter with four different optimization
> > levels.  The compiler switches used were '-G' (no optimization), '-O1,
> > '-O2', and '-O3'.  bogofilter was run 1000 times on a test message.
> >
> >                 g       O1      O2      O3
> > user          18.17   15.69   15.05   14.89
> > system        3.24    3.40    3.17    3.11
> > elapsed       21.44   19.32   18.23   18.00
> >
> > It appears that maximum speed is about 1/7 (14%) better than worst code.
> >
> > Notes: tests run on 500 mhz Pentium III with 256mb ram, kernel
> > linux-2.4.19-16mdk.
>
>On reading the above, I decided "just for kicks" to try my
>0.8.0-derived production version with -O3.  Variation from run to run
>of the "time" command is about 2% so it's hard to be sure, but any two
>adjacent runs (alternating between -O2 and -O3) show the -O3 version
>(./bogofilter) being about 35-50 microseconds (~1%) faster:
>
># time for i in `seq 1 2000`; do bogofilter </root/testmsg; done
>
>real 0m15.290s    user 0m10.650s    sys 0m4.610s
>
># time for i in `seq 1 2000`; do ./bogofilter </root/testmsg; done
>
>real 0m15.218s    user 0m10.590s    sys 0m4.600s

Given you're running 2000 reps in 15 sec, vs my 1000 in 18, you're running 
bogofilter about 2.4 times faster than I am. Interesting.


>That's on a 1.1 GHz (500 millihertz is a bit slow, no? :) PIII/mobile with
>512Mb RAM / kernel 2.4.20-rc2-ac3.

When I got the machine 3 years ago, the state of the art was about 2 speed 
steps faster than 500 mhz, which made the cost/performance ration of 500 
mhz very attractive.  I've had no need since then to upgrade, though some 
of the large bogofilter tests of the last 2 months have got me thinking 
about upgrading to a dual athlon 1.0+ ghz.  However finances are tight and 
I can't justify that.