> >Just wandering around the kernel, and it seems that if "COMPAT_42" is defined
> >(which is everywhere), then UDP checksumming is switched off.  This seems a
> >Bad Thing (tm), especially on this wonderful Internet of ours :-(
> >
> >What would happen if it were turned on by default in the distribution?  Would
> >NFS stop working to Suns, or will it ignore the checksum if the field is 0?

The Sun's at least seem to ignore checksums if 0 even if turned on...

> SunOS 4.1.x Suns have UDP checksums turned off by default, so it's not very
> likely that NFS will stop working to them (-:

We've turned on UDP checksums on our SLIP gateway host without turning them
on everywhere else with no ill effects.
 
> I usually turn them on in my home SPARCstation-1 clone and on a target Sun if
> I expect to do NFS over my PPP link, but otherwise I live happily without 'em.

BTW...we've seen under 4.1.x that if you're streaming data at a high
rate (14K and up) out of the Sun4c serial ports or even the
multi-serial/parallel I/O board, that it munges the data about every
15-16k regardless of dinking with window and packet sizes.  The munging
is somewhere in the tty or zs drivers.  It's not getting munged by the
link (v.42 and MNP layers have no complaints), but it's coming out of
the serial port munged already.  Note that going into it we don't have
a problem (other than low throughput b/o lousy Sun serial port
design...too many interrupts and other problems).

On the otherhand, NetBSD has no problems like this...at least that I've
seen.  Talking to other boxes with better serial/tty hardware and drivers
I'm able to get 80-90% of theoretical throughput...even with V.fast at
28.8k.
--
-Mark

------------------------------------------------------------------------------