buildworld times

[Bruce Evans]

On Sat, 28 Feb 2004, Don Lewis wrote:

> On 28 Feb, Vincent Poy wrote:
> > 	I was wondering what people are getting as the fastest make
> > buildworld times since it seems like a P4m-2.6Ghz 2GB RAM, 7200RPM HDD
> > machine when using make -j4 buildworld or just make buildworld takes 2.5
> > hours.  I guess I'm trying to look for what the average times for
> > buildworld is supposed to be for a Pentium 4 Northwood based platform.
>
> This seems way too slow to me.
>
> My Athlon XP 1900+, 1GB RAM, 10K RPM SCSI drive machine running
> 5.2-CURRENT with the kernel options:
> 	SCHED_ULE
> 	INVARIANTS
> 	INVARIANT_SUPPORT
> 	WITNESS
> 	WITNESS_SKIPSPIN
> 	DEBUG_VFS_LOCKS
> and no /etc/malloc.conf runs buildworld in
> 2563.925u 1977.980s 2:01:04.11 62.5%    2715+1768k 20833+2389io 6587pf+0w
> while running setiathome.

This seems way too slow to me :-).  Well, maybe not with all those
debugging flags.

My Athlon XP 1600 overclocked by 146/133, 512MB RAM, 7200 RPM IDE drive
machine ran makeworld in 1390 seconds when it was new 2.4 years ago:

%%%
Athlon1532
--------------------------------------------------------------
>>> elf make world completed on Fri Jan  4 00:00:15 EST 2002
                    (started on Thu Jan  3 23:37:04 EST 2002)
--------------------------------------------------------------
     1390.18 real       913.56 user       232.63 sys
   ...
   9055054  page reclaims
     21976  page faults
         0  swaps
     31885  block input operations
      4960  block output operations
   ...
    346041  voluntary context switches
    423794  involuntary context switches
%%%

My most recent complete buildworld in a (mostly) controlled environment
took 2788 seconds last November:

%%%
bde-current with ... + KSEIII + idlezero_enable + pmap - even coloring
async mounted /c
sources in /cbak/src
no /etc/make.conf
after perl removal and new gcc and ufs2 and aout utilities removal
after kerberos5 addition
after rescue addition
with 2 fairly new drives
1532 MHz AthlonXP 1600
1024MB
make catches SIGCHLD + par-all
i686_bzero used and replaced by one that uses SSE (movntps)
kernel up 8 days
--------------------------------------------------------------
>>> make world completed on Tue Nov 11 21:55:53 EST 2003
                   (started Tue Nov 11 21:09:25 EST 2003)
--------------------------------------------------------------
     2788.41 real      2316.49 user       357.34 sys
  ...
  18994831  page reclaims
     13700  page faults
         0  swaps
     22756  block input operations
      7747  block output operations
  ...
    289983  voluntary context switches
    809112  involuntary context switches
%%%

Normally I reboot to run benchmarks, but the above just ensures not
getting any local _source_ optimizations by building a freshly checked
out copy of the src tree and no /etc/make.conf.

The machine now has 1024MB of memory and faster drives.  The build
time was increased mainly by gcc compile-time pessimizations (80%
of the user time) and more things to compile (50% of everything?)
and breakage of not building dependencies (5-10% of overything).
The larger RAM and fast drives only compensate for 5-10% of the
bloat.

Bruce