[TUHS] Attempting To Build NOSC and BBN UNIXs + ARPANET code
pnr at planet.nl
Wed Oct 12 01:08:15 AEST 2022
>> The BBN with TCP stack is a bit mislabeled: it still appears to
>> support NCP, but none of the client apps are there, but its directly
>> built off the NOSC stack.
> That's very good. I hope the NCP support there is in good shape.
>> it's probably a fork from earlier in development. 79-80 timespawn
>> would have been *very* early in TCP's life
> TCP had been underway since 1973. Experiments called "TCP bakeoffs"
> started around 1979.
That is what the “V6 with TCP” on TUHS is:
Following the success of NCP Unix, it became a base for various TCP experiments in the ’77-’79 era. The first was an implementation by Jack Haverty, that wrapped an existing TCPv2 stack that was written in PDP-11 assembler into a Unix application. It ran in user mode and depended on Rand ports and several extensions that Jack added to the kernel (await/capac and a user mode timing variable, where the clock routine incremented a variable in user space). He used a PDP-11 with little core and the pipes (ports) did not stay in the file buffers, but flushed onto disk. This killed performance: Jack recalls that a bad run would average a few characters per second.
Next Mike Wingfield wrote a TCPv4 stack in C, more or less using the architecture of the above. It was the “winner" of the December 1979 bake-off. I think it is the first TCPv4 implementation for Unix and maybe the oldest surviving source for TCPv4 overall. I wanted to try if it would still interoperate with modern TCP/IP, but I never got around to that. An actual printout survives in the SRI archives and I painstakingly retyped that source, just weeks before Noel found the right tapes :^). Later still, Craig Partridge found a full report and listing in the BBN archives (report no. 3724). This NCP Unix with the Wingfield library is the version that is labeled “BBN V6 with TCP” on TUHS.
Some of the code in the Wingfield stack is to test the protocol. Arpanet essentially offers circuit switching, and some of the code is there to simulate dropped packets, out-of-order packets, etc. It also tested security features that were under consideration, but subsequently dropped as interest shifted to end-to-end encryption.
Again, user mode TCP was not found to be practical, the 16-bit era was ending and that is when Rob Gurwitz was assigned to write a new stack for the VAX (1980). By that time Jack Haverty was his boss. Some parts of the BBN VAX-TCP design still echo the user space origins and experiences of the BBN team in the immediate years before. This stack I got working and it still interoperates with modern TCP/IP (at least it did some 5 years ago).
Jack Haverty can easily be reached via the internet-history mailing list.
I’ve summarised the history here: http://chiselapp.com/user/pnr/repository/TUHS_wiki/wiki?name=early_networking
I should transfer it to the TUHS wiki or to Gunkies one of these days.
I am not sure I understood which files are missing or corrupted and in which NCP Unix trees. Noel retrieved the files from old (mouldy even) tapes so some corruption is quite possible. Pulling together material from various sources can hopefully lead to a working source tree. Glad to help.
Further note that NCP Unix was initially developed on 5th Edition, but soon migrated to 6th edition. I am sure that the various installations tracked new developments and installed extra bits and pieces. The surviving images are from 1979 and for sure would have picked up bits from newer releases and other sources (such as the Uni of Calgary buffer extensions).
More information about the TUHS