[HECnet] TCP/IP experts?

[Johnny Billquist]

On 2013-02-17 02:41, Clem Cole wrote:
On Sat, Feb 16, 2013 at 8:12 PM, Johnny Billquist <bqt at softjar.se
<mailto:bqt at softjar.se>> wrote:


      I honestly don't know for sure, but this is the first I've ever
      heard of that.
      Doing a "portable" stack for RSX would probably be very hard, by the
      way.

Actually doing a portable stack was hard for any OS, but that was idea.

True. Portable do have its advantages. But it always comes at a price, and for something like RSX, that price is probably way too high. There are big discrepancies in the I/O model, memory management as well as process handling and god knows what else...

  It was written originally in C by Rob Gurweitz and some other ??Bob
Walsh maybe?? - I want to say in 1978/79 time frame, using the original
Ritchie compiler.     There were a ton of tools developed in the UNIX
community to move code at the assembler lever from V6/V7 to RT11 & RSX.
  Look at the old Harvard tape that I think Warren has in his archives
if you are interested.   I do not remember how it moved the HP's, but
there must have been an 3000 or 1000 hack for the Ritchie compiler at HP
and/or Berkeley.

C would definitely not been an option at that point. Not in the RSX world. I don't even think DECUS C had been published, and that compiler have some serious limitations.

As for the BBN portable IP/TCP stack, it is where the famous "mbuf"
structure came from.   Rob needed an OS independent memory system for
incoming packets.   The BBN IP/TCP was written to call it, and then on a
kernel for kernel basis the mbuf system was written for the native OS.

mbufs makes a lot of sense. I did something similar for my TCP/IP.

      ...
      you want really bad performance, you want to be close to the kernel,

That was not an goal.     You have to remember that at the time of IP's
development, every computer firm had it's own proprietary networking
scheme.     DARPA wanted something that worked on all them, so they paid
to have it created and implemented.     But it also meant that DARPA
needed those implementations of IP/TCP for the system that their primary
folks were using - the PDP-10, Multics, UNIX.   In some labs (like US
Army's BRL if IIRC), there were realtime systems doing specific tasks.
  HP and DEC were the two firms selling those systems.     So BBN needed
an implementation of IP that >>worked<< that allowed those systems to
connect.

Right. But thinking a bit more, I'm pretty sure there was no TCP/IP for RSX. You even have all those old RFCs with comments and lists about systems available, and implementations... I've been over them a few times. I'm pretty sure RSX is not in there.

When we finished it, some of my friends/former colleagues at CMU/CS
asked me for it, we made it available (no stinking GPL, or calling it
"open source"   - we just shared it).     They wrote the mail 11 interface
and cleaned the whole thing up a great deal.

Open source and free software is not what it used to be. I know... :-(

	Johnny

-- 
Johnny Billquist                                   || "I'm on a bus
                                                                  ||   on a psychedelic trip
email: bqt at softjar.se                         ||   Reading murder books
pdp is alive!                                         ||   tryin' to stay hip" - B. Idol