unimpressive buildworld time

Garrett Cooper youshi10 at u.washington.edu
Thu Nov 15 21:35:30 PST 2007

Tim Daneliuk wrote:
> Jonathan Horne wrote:
>> On Wednesday 14 November 2007 03:45:07 pm Aryeh M. Friedman wrote:
>>>> Impressive ;-) My main machine (with an Athlon XP @ 2GHz) takes ~2
>>>> hours to build kernel and world (I use a script to do that). My
>>>> other box is running -CURRENT and takes ~11 hours to build kernel
>>>> and world (Celeron 500...).
>>>> Just to supply some numbers that "go the other direction" :-)
>>> With no -j and running gnome and doing other things in the foreground
>>> (watching a avi) 1 hr 3 mins on a e6850 w/ 4 gig (amd64)
>> p4 540 3.2GHz, 1GB ram:
>> --------------------------------------------------------------
>>>>> World build completed on Thu Nov 15 19:15:05 CST 2007
>> --------------------------------------------------------------
>> real    63m8.635s
>> user    102m44.096s
>> sys     10m44.889s
>> [root at athena /usr/src]#
>> heh, i have appropriately renamed the thread.  :)  with -j 8
>> cheers,
> My Pentium-D 3G DualCore w/2G memory and a pretty vanilla
> SATA drive does buildworld and 3 different kernels in 68 minutes wall 
> time
> building 6.3-PRE (aka -STABLE) using -j20.

SMP kernels on STABLE (6.x) are going to perform worse than SMP kernels 
on CURRENT (7-RELENG / 8-CURRENT), depending on the scheduler used (4BSD 
vs ULE scheduler), as well as a variety of other factors.

Remember... performance not only depends upon clock speed or the number 
of cores you have, but also what caching/prefetching scheme FreeBSD uses 
(not sure if it's fetches large amounts infrequently or small amounts 
frequently), how much memory is available to make and its spawned 
processes (gcc, awk, etc), as well as the number of processes active on 
the machine, and host usage (high disk usage, high memory usage, etc).

After reading through the thread, I noticed that people are making 
comparing apples to oranges, as...

1. 6.x contains gcc 3.4.x whereas 7.x/8.x contains 4.2.1
2. The machines have a variety of different resources available.
    - RAM types and amounts.
    - Processor types (AMD processors typically have lower cache sizes 
than Intel processors).
    - Different disk types and speeds.
    - etc, etc.
3. Load amounts vary.

If you would like to run realistic benchmarks and money isn't an issue, 
I suggest buying similar (if not same) hardware and experimenting with 
only one variable item (i.e. harddrive, processor, RAM, etc), or using 
one machine as a testbed, varying hardware components and software 
variables to yield repeatable results, instead of just saying, "Hey, my 
make buildworld is slow even though I have an uber off-the-market 
processor / setup -- what the heck man???", or another similar claim.

My 2 cents..

More information about the freebsd-questions mailing list