[ih] When did "32" bits for IP register as "not enough"?

[Bob Hinden]

Craig,


> On Feb 16, 2019, at 8:35 AM, Craig Partridge <craig at tereschau.net> wrote:
> 
> Hi Bob:
> 
> You are right about the implementations being in parallel for a while.  I'm going to get the dates slightly wrong but Bill's code came out in 4.1c (1982?) and BBN kept maintaining its implementation (Dennis Rockwell and then Bob Walsh) until c. 1985.  Indeed, as I recall there was a debate c. 1985 at DARPA about whether to force Berkeley to use the BBN code (which had been modified to use the sockets API, which everyone agreed was better -- I had a small part in the port to sockets).

I was doing the TCP/IP implementation at BBN for the TAC around the same time that the BBN Unix TCP/IP was being developed.  I suspect Rob and I did some testing of our implementations, but don’t remember any details.  I do remember having to deal with doing 32-bit sequence number arithmetic on a 15 bit machine :-)

Thanks,
Bob

> 
> I believe that when the TCP/IP project transitioned to me (c. 1987?), I stopped effort on the BBN TCP/IP.  I think we had to maintain it a bit as we'd licensed it to a workstation vendor (Apollo????).  Karen Lam and David Waitzman switched to the BSD TCP/IP and we did things like help Steve Deering implement multicast and such.
> 
> Thanks!
> 
> Craig
> 
> 
> 
> On Sat, Feb 16, 2019 at 9:24 AM Bob Hinden <bob.hinden at gmail.com> wrote:
> Craig,
> 
> > On Feb 15, 2019, at 12:38 PM, Craig Partridge <craig at tereschau.net> wrote:
> > 
> > Important historical nit.  I was the manager of the BBN UNIX TCP/IP effort after Rob Gurwitz left (I think Rob inherited it from Jack Haverty, but not sure).  The BSD stack with sockets *was not written by BBN*.  It was written by Bill Joy at Berkeley -- using the earlier BBN 4BSD code as a reference.  Entirely new code, but originally bug-for-bug compatible (indeed, years later, when a bug was found in the BSD TCP, the BSD folks stood up and said "that's a bug from BBN”)
> 
> My memory is that they both were maintained in parallel for several years.  Also, Bill Joy's TCP stack also had the “trailers” feature where the headers were at the end of the packet.   Documented in RFC893 (https://tools.ietf.org/html/rfc893).   Probably not so good for packet switching, but better performance on some host implementations.  That faded away at some point.
> 
> Bob
> 
> 
> 
> -- 
> *****
> Craig Partridge's email account for professional society activities and mailing lists.