Howdy, Occasionally the question pops up on the questions@ list about what the fastest -jX number is for a single CPU system. I had some spare time so I tried out a small matrix of possibilities.
My conclusion is that using -jX at all is mostly a waste of time on
single CPU systems running -STABLE (even with multiple spindles being
involved), especially when one considers that -jX may introduce build
problems.
NOTES:
* I used the simple shell "time" command
* /usr/obj was cleaned out before each run and I waited at least 30
seconds afterwards for write caching to settle down
* This is my regular build host for my network
* I built 4 kernels: 3 customized and GENERIC (see above for why)
* Celeron 900, 256Mb of RAM, /usr/src and /usr/obj are both on their own
set of spindles
* /usr/obj consumes part of a vinum mirror on dual 40Gb 7200RPM
Maxtor 6L040J2's (the remaining vinum filesystems weren't active
during this test)
* /usr/src is on a 2,1Gb Compaq ST32550N SCSI-2 drive
* The operating system is on separate spindles
RESULTS:
buildworld -j2 buildworld -j3 buildworld -j4 buildworld
========== ============== ============== ==============
real 57m10.367s 54m10.992s 55m7.494s 55m1.459s
user 38m5.436s 38m20.852s 38m22.453s 38m23.056s
sys 9m2.801s 10m12.876s 10m17.140s 10m14.792s
buildkernel -j2 buildkernel -j3 buildkernel -j4 buildkernel
=========== =============== =============== ===============
real 36m59.994s 36m58.988s 37m42.956s 37m31.627s
user 29m35.597s 29m43.405s 29m43.846s 29m48.652s
sys 4m50.478s 5m26.372s 5m26.883s 5m22.763s
Thought this might be of some interest,
-T
--
Re: alt.sysadmin.recovery
A fitting punishment for kindly naivete, to end up belonging here.
- A.S.R. quote (Chris Johnson)
pgp00000.pgp
Description: PGP signature
