[ih] error detection

[Steve Crocker]

On Thu, Oct 1, 2020 at 6:22 PM Lawrence Stewart via Internet-history <
internet-history at elists.isoc.org> wrote:

>
> There was a time, I think that people thought the hardware link checksum
> was sufficient, and indeed the CRC-32 is much better than the TCP software
> checksum.
>
> Folks quickly realized that important parts of the overall system were not
> protected.  Datapath inside adapters, memory bus transfers, bad memory in
> hosts and routers, etc.
>
> Consequently an end to end checksum is essential.



It happened even more quickly than you're suggesting.  In mid-February 1969
a few of us in the Network Working Group met with the BBN group for the
first time.  This was about six weeks after they had begun work on the IMP
contract and 6-1/2 months before the first IMP was delivered to UCLA.  We
had a rough cut at the host-host protocol in mind.  We intended to
include a lightweight end-to-end checksum.  Sixteen bit ones complement
(end around carry) with one bit rotation every approximately thousand
bits.  Jeff Rulifson had argued this was good practice and might catch
implementation errors at different layers.  The one bit rotation was a
small wrinkle we included to catch possible errors in reassembly of packets
into messages.  We knew the communication between IMPs would be protected
by 24 bit hardware checksums; we weren't trying to complete with that.

Frank Heart reacted strongly.  "You'll make my network look slow!"

We went back and forth briefly.  I pointed out the path between the host
and IMP was not protected and asked how reliable it would be.  "As reliable
as your accumulator," insisted Heart.  To my regret, we -- primarily I --
relented and didn't include a checksum in the original host-host protocol.
Sure enough, there was a problem with the early host-IMP connection on the
Lincoln TX-2 -- harmonic interference between the drum and the host-IMP
interface when both were operating, I believe.  I heard they had a devil of
a time tracking it down.

Steve


>
>
>