ponderous 'make world' times post GCC 3.3...

[Garance A Drosihn]

At 11:25 PM -0700 8/6/03, David O'Brien wrote:
>Am I the only one that saw 'make world' go from almost 3 hours
>with GCC 3.2 to: [...]   11h    17.00s real
>                           9h29m 42.85s user
>                           1h15m 22.05s sys
>
>post GCC 3.3?

I'm not complete done yet, but here is my observation:

Back on July 6th, I did a 'time make -j2 buildworld' on my
Ultra-10.  Let's call that build #1.  Last night, I did the
same command (after cvsup-ing my source tree).  I started
that build on the very 'world' installed on July 6th.  Let's
call this build #2.

              Build #1         Build #2
            ------------     ------------
    real    347m 33.407s     387m  4.205s
    user    283m  0.162s     315m 25.441s
    sys      53m 45.805s      59m 44.648s

So, yes, that's a little slower.  It isn't horrendous, though.
And there is a good chance that some of that slowdown is due
to non-gcc factors, such as building /rescue (which was not
being built in July).

I then installed that 'world'.  After installing that world, I
removed /usr/obj/usr/src/*, I did *not* cvsup, and I built the
exact same world again.  This is build #3.  Note that I am sure
I had also removed /usr/obj/usr/src/* before build #2, and it
is pretty likely that I did before I started build #1, too.

As I write this message, build #3 has been running for over
fourteen hours, and based on the logfile size I would say it
isa less than 6/7th's of the way through.

So, apparently the problem is something a bit more subtle than
just gcc 3.3 being slower to compile than gcc 3.2.  Apparently
the August 9th system is a lot slower at *running* than the
early system.  Do we have some other benchmarks we could run?

I would also note that the CPU seems pinned at 0% idle, and I
do not believe that used to be the case when I was doing a
'make -j2 buildworld' on this system.  By "pinned" I mean
every single update of 'top' for the past 20 minutes has
showed 0.0% idle.  At no time has it been even 0.1% idle.
This reminds me of how my older i386 box used to run when
WITNESS was turned on in the early days of 5.x-current!

My kernel config is almost exactly GENERIC, except that I turn
off WITNESS* and SMP.  The machine is not swapping at all (it
has 384 meg of RAM, it does not have X11, and really the only
thing it's doing is the buildworld).

I'm not sure what else to say which might be of some help.

-- 
Garance Alistair Drosehn            =   gad at gilead.netel.rpi.edu
Senior Systems Programmer           or  gad at freebsd.org
Rensselaer Polytechnic Institute    or  drosih at rpi.edu