Subject: Re: Compiler timings on varous MVII NetBSDs etc.
To: James Lothian <>
From: NetBSD Bob <>
List: port-vax
Date: 01/25/2001 14:29:06
> > > > But you're not thinking about running NetBSD 1.5 on the pdp11 ?

Awww, heck, why not.....(:+}}... I know torture....

> > > Interesting.  My take on it, is that like the PDP11 and the 4M ram
> > > limitations, we are reaching a 16mb ram limitation in the old VAXentoyz.
> > > I am sensing that gcc may be running out of ram.  Especially in the
> > > 25 hour kernel compile arena and the neverending perl compile thicket.
> > > Isildur got a kernel compile in 30 minutes on a KA650 box.
> > 
> > Only the MVII, MS2000 have the 16mb limit.  LAter machines are at least
> > capable of 32mb or much more.
> > 
> > personally I find it ghastly to need more than 16mb to compile reasonably.
> > Sound much more like disk flogging (not enough or large enough IO
> > buffers?).
> Erm, the 750 has a 14Mb limit. And most 750s were 8Mb. I know it's not a
> microvax, but it's still a vaxentoy, at least as far as I'm concerned!
> And you can build a 4.3 kernel on it in about a couple of hours. 

When I picked myself up off the floor after my MVII with 7mb ram,
running Tahoe had finished compiling its kernel in 45 minutes, and
the NetBSD-1.5 took 25 hours, I figure something must not be quite
right somewhere.  To ferret out that where, is the question....

4.3BSD        0.75 hr  (7mb ram)      273K
NetBSD-1.2    3 hr     (13mb ram)     315K
NetBSD-1.3    ?            "          360k
NetBSD-1.4.3  5 hr         "          470K
NetBSD-1.5    25 hr        "          580K

These are all compiled on a stock stripped MVII kernel, using
a stock stripped MVII config, basically all were identical to
the 4.3BSD config.

Something has got to be happening.  Everyone says its  ``features''.
I still can't accept that as the entire answer.  I smell some gcc
things going on, but I can't lay my finger on it.

Offhand, can I compile a NetBSD-1.2 kernel on NetBSD-1.5
by unrolling into a separate tree, and not break anything?
My thought would be to compile the identical 1.2 system
stripped MVII kernel on the 1.5 box, and see how big it
became.  If the kernel was approximately the same size, no
problem.  IFF it grew significantly, that might indicate
gcc was having problems.  I may have to try that on my M76.

Although, Isildur has that fast KA650 box that does a 30
minute compile.....(:+}}... hint, hint.

Yeah, I know it is heresy to throw bones at gcc, but my
closet has lots of skeletons.....(:+}}, and I gotta get
to the bottom of the pile, somehow.