numbers don't lie ...
danny at cs.huji.ac.il
Thu Sep 21 22:59:53 PDT 2006
> Danny Braniss wrote:
> > you might have a point, but this started when I asked why, two
> > boxes, under similar test gave idential real times, but very different
> > user times.
> Right, and the answer was: One box has a much faster CPU,
> so it's user time is smaller, but buildworld isn't purely
> CPU-bound, and because of I/O delays the real times end
> up to be about the same. In other words: The faster box
> had to wait more often for the disk than the slower box.
> If both of your machines have enough RAM, it would be
> interesting to repeat the test with /usr/src being in a
> RAM disk, so read I/O doesn't play that much of a role.
> Best regards
> PS: Numbers don't lie ... but are often misinterpreted.
or missused by salesmen/politician/etc :-)
i have run many tests, even having /usr/src in memory, the
results where posted some days ago, but here they are again:
make: dell 2950
OS: Freebsd 6.2-PRERELEASE
cpu: XEON 3.20GHz dualcore * 2
src & obj real user system hyper
-------------------- --------- ---------- --------- -----
Dell PERC 5/i RAID 0 24m17.73s 1h4m31.49s 15m47.44s no
Dell PERC 5/i RAID 0 22m3.39s 1h38m46.84s 28m54.18s yes
iSCSI/netapp 26m49.98s 1h4m26.77s 16m12.89s no
md Dell PERC 5/i 24m7.22s 1h4m44.94s 16m24.45 no
so something is still fishy in the state of Denmark.
or, in the case of this box, the cpu is so slow, that no matter how fast the I/O is
is does not change the equation.
I will try to run similar tests on the amd/sun, but have to wait till some
real work finishes.
More information about the freebsd-hackers