VM config managment problems...

Ian Piumarta ian.piumarta at inria.fr
Wed Sep 10 15:04:43 UTC 2003


On Wed, 10 Sep 2003, John M McIntosh wrote:

> > Hands up anyone running Unix on a 68k machine...?
> Wait wait there is hand here in the corner. I've got TWO OpenBsd 68K
> boxes running. And I know Bruce ONeel was running a 68K NetBSD box

Ah, okay.  I think 3 counts as a quorum. ;)

So what are the benchmark results with/out the jump table pointer
modification on all of the above machines?

> > Even better, check benchmark times.  Removing instructions has been
> > known to make things slower.  (Too many hazards in the pipeline
> > end at the same time.)
> Mm that is interesting, have you observed that on the powerpc?

Yes, but only on older (600 series) cpus -- if I remember correctly.  
(There's something about this in the architecture manual too; if you
_really_ want I can try to dig up the ref.)  On the 700s my experience is
that you reach a critical density above which removing instructions makes
no measurable difference in speed (which I think is what we saw when we
tried the code on my G3).  I don't know how the 7000s are supposed to
behave, but I'm sure it's no worse than the 700s.  (Is Tim O reading
this?? ;)


Cheers,
Ian





More information about the Squeak-dev mailing list