[ih] Fwd: early networking: "the solution"

[Craig Partridge]

On Mon, Apr 22, 2024 at 10:21 AM Greg Skinner via Internet-history
<internet-history at elists.isoc.org> wrote:

> >
> > I have vague recollections the Internet checksum pointed out a bug in the SINCGARs applique prototype during development.  We were  getting packets that had IP checksum errors  yet had passed all the lower layer checks in the radio portion.  That wasn't supposed to happen.

A similar, if perhaps more horrifying situation.  Sun's Network File
System (NFS) initially shipped with the UDP checksum turned off -- it
made file system performance over a local Ethernet enough faster that
it mattered competitively.    But this meant, of course, a trashed UDP
packet* could trash your file system.  And, for years, trashed NFS
file systems were a fact of life...

*And there were multiple trashing-capable components of the network
path, ranging from the Ethernet switches, to the network interface
cards, to the system bus.  Indeed, I know one computer vendor who had
a super rare bus error in which bits got trashed -- they couldn't find
it, but knew it existed because their NFS file system would fail every
few months.  So they turned on the UDP checksum in NFS, the problem
"went away", and they could ship their product.

Craig
-- 
*****
Craig Partridge's email account for professional society activities
and mailing lists.