Re: pmax & NetBSD 4.0

On Wed, 17 Oct 2007 11:47:12 +0100 (BST)
David Brownlee <[email protected]> wrote:

> On Tue, 16 Oct 2007, George Harvey wrote:
> 
> > On Tue, 16 Oct 2007 14:27:06 +0100 (BST)
> > David Brownlee <[email protected]> wrote:
> >
> >>  	So, what are people doing with their pmaxen, and is anyone
> >>  	planning on testing the NetBSD 4.0RC3 build which should be
> >>  	up at ftp://ftp.netbsd.org/pub/NetBSD/NetBSD-4.0_RC3/ later
> >>  	today? :)

[ snip ]

> > as normal. All my Turbochannel cards are recognised (PMAGB-B,
> > PMAD-A, DEFTA-AA) but I haven't checked if the PMAD and DEFTA work
> > yet. I'm expecting GCC 4 to be slow so I'm running a test at the
> > moment to see how it compares with GCC 3 on 3.1.
> 
>  	Short answer, uses significantly more CPU and memory to compile,
>  	produces very slightly better code :/ Its a fine tradeoff if you
>  	have a fast box or can crosscompile from one, but less fun if
>  	you need to natively compile on a slow machine :(

As expected, it's slow. Just for info, here are the times to compile Tcl
8.4.11 on my 5000/133 with different NetBSD/GCC versions, these are all
on a local SCSI disk:

  NetBSD  GCC          sys        user        real
  1.4.3   1.1.1     6m57.0     38m38.9     45m24.8
  3.1     3.3.3    10m31.6     78m38.0     89m37.3
  4.0RC2  4.1.2    10m32.1    105m16.1    116m25.1

>  	Compiler aside it would be interesting to see how the machine
>  	compares to NetBSD 3 in terms of performance and stability.
>  	the problem is that would involve having a NetBSD 3 and NetBSD 4
>  	installs available and switching between them to run something
>  	like ttcp, bonnie, and apachebench... (and plenty of time :)

You would certainly need plenty of time! I tried compiling bash from
pkgsrc NFS mounted over the FDDI interface. It worked, but it was very
slow. As an alternative, I've used my RaQ2 (also mipsel) to do native
builds of a few packages then installed the binaries on my 5000. That
works a bit quicker and I now have afterstep 1.8 as my window manager. X
seems to work pretty well on the PMAGB-B but I did get a couple of seg
faults when trying to run the Tk 'make test' suite. These don't look
like they're pmax specific though, since they also happen when I run the
same test on the RaQ2.

George