Re: buildworld times

On Mon, 1 Mar 2004, Vincent Poy wrote:

> 	After taking out -WITNESS in the kernel config, I'm getting 1
> hour and 13 minutes and 50 seconds, a bit improvement:  3141.937u
> 1093.463s 1:13:50.47 95.5% 2902+1955k 18436+2525io 6512pf+0w
> 
> Pentium-4 Mobile 2.6Ghz 400FSB, 2GB DDR266 memory, Hitachi TravelStar
> 7200RPM 60GB 2.5" HDD.  Temps do hit 87.7C though, probably due to the
> way notebook cooling is on a Dell Inspiron 8200 and the heat from the HD
> and the Nvidia Quadro 4Go 700GL. 

FYI, I now have access to a build box at work with two Xeon 2.4GHz
processors, each with two logical CPUs, and 1GB of memory.  Here are the
buildworld times, with -DNORESCUE and -DNOPROFILE, 5.2.1-RELEASE
GENERICish kernel (no WITNESS, INVARIANTS):

		Real		User		Sys
default		2195.16		1717.69		467.78
-j 2		2003.20		2151.49		539.67
-j 4		1703.15		2485.99		654.00
-j 6		1645.34		2595.67		718.12
-j 8		1627.88		2618.15		743.53

I assume this is largely a product of hyperthreads (less CPU for the same
user time, requiring more user time to get the same amount of work done),
but it's interesting how the usertime goes up substantially with
parallelism.  I imagine these numbers would be somewhat better with
5-CURRENT due to performance improvements there; they'd also be slightly
better if I unset the debugging flags with malloc.conf.

For better or for worse, the 5.x source tree doesn't compile with earlier
(faster) gcc versions.  Statistics posted previously suggests that we paid
a substantial price to move to gcc3, although hopefully that will improve
with time.  There was recently a somewhat heated discussion on the openbsd
tech or misc mailing list on the topic of gcc compile time pessimization
in which the GCC developers seemed sensitive to the issue, albeit somewhat
on the defensive.

Robert N M Watson             FreeBSD Core Team, TrustedBSD Projects
robert_at_fledge.watson.org      Senior Research Scientist, McAfee Research