[nycbug-talk] make -jN performance

[Tillman Hodgson]

On Thu, Jan 20, 2005 at 11:19:11AM -0600, Tillman Hodgson wrote:
> On Thu, Jan 20, 2005 at 12:14:08PM -0500, dlavigne6 at sympatico.ca wrote:
> > Is this on a 4.x or a 5.x system? I'm wondering if there's a difference between the two...
> 
> In my case it was a 4.x system (following -stable at the time). I can
> try it on a -current system if you don't mind waiting a while for
> results.

And I've now done such tests on a -current system.

# uname -a
FreeBSD backforty.seekingfire.prv 6.0-CURRENT FreeBSD 6.0-CURRENT #2: Fri Nov 19 08:03:52 CST 2004 tillman at backforty.seekingfire.prv:/usr/obj/usr/src/sys/BACKFORTY  i386

# dmesg <-- I snipped out the interesting bits of the output
CPU: AMD Athlon(tm) XP 1700+ (1477.51-MHz 686-class CPU)
real memory  = 268369920 (255 MB)
avail memory = 253235200 (241 MB)
ad0: 19092MB <WDC WD200BB-00AUA1/18.20D18> [38792/16/63] at ata0-master UDMA100

(Yes, a single drive, so /usr/src, /usr/obj, and the OS share a single
spindle).

# cd /usr/obj && rm -rf usr && sleep 60 && cd /usr/src
# time make buildworld
real    78m2.997s
user    62m33.527s
sys     7m48.909s
# cd /usr/obj && rm -rf usr && sleep 60 && cd /usr/src
# time make -j2 buildworld
real    81m7.560s
user    65m47.179s
sys     8m45.934s
# cd /usr/obj && rm -rf usr && sleep 60 && cd /usr/src
# time make -j3 buildworld
real    81m47.648s
user    67m25.276s
sys     8m59.892s
# cd /usr/obj && rm -rf usr && sleep 60 && cd /usr/src
# time make -j4 buildworld
real    84m54.193s
user    70m3.855s
sys     9m21.527s

On this particular hardware, the results got progressively *worse*.

-T


-- 
The correct way to punctuate a sentence that starts: "Of course it is none of
my business but--" is to place a period after the word "but." Don't use
excessive force in supplying such moron with a period. Cutting his throat is
only a momentary pleasure and is bound to get you talked about.
    -- Robert Heinlein