Ross Harvey wrote:
> On Tue, 20 Jul 1999 18:51:57 +0200 (MET DST) 
>  Joerg Czeranski <jc@joerch.org> wrote:
> 
>  > I ran a mildly memory intensive integer benchmark (Ford/Fulkerson
>  > network flow optimization) on a PC164, and it took 50.3s with
>  > NetBSD-1.3.3 and 48.4s with DU4.0E, same assembler source, but assembled
>  > and linked on each target OS.
> 
> I would be interested in running it as well. Did you profile
> this program on both systems?  I wonder if a library call could be
> a factor.

I profiled it on a AXPpci33 under DU4.0D.  > 80% are spent in three
rather short functions, almost all the rest in the garbage collector
(it's written in Objective Caml, a functional language).

> Are these results consistently reproducible?  I suspect DU4.0 will be
> using page coloring...

Yes, they are, and yes, I think DU4 does.

> Also, how big is this binary image?  We don't at the moment deal with
> a certain ELF performance problem that happens in very large programs.

The important code should fit into an 8KB I-cache.  The complete
binary is text = 131072, data = 24576, bss = 272 for DU, and it's linked
to libc, libm and libcurses, but doesn't really use them much.

It's a rather silly benchmark, BTW, but if it helps optimize NetBSD
performance, who cares? :)

I'll be offline for the rest of this week, but I'll get back to it
next week and make source and binaries available.

joerch