On Fri, Jul 16, 2021 at 3:08 PM Kevin Bowling <kevin.bowling@kev009.com>
wrote:

> Yup was just going to say this is standard in the modern BSD network
> drivers, looks like Clem says it's older.

Absolutely -- I believe it was Rob's undergrad project at Brown that he
brought to BBN.

The first use, if I saw, was the 'portable IP/TCP' stack  BBN did for
HP/3000 and a couple of other systems.  That code seems to have been lost.
I have asked about it on the Internet history mailing list.  I had a copy
of it one time, but sadly I don't think I still do.  IIRC The original
PDP-11 IP implementation which ran on a couple of dedicated systems,
whose names/function I frankly do not remember) was also based on a version
of this code.  I think it ran something like RT-11 or DOS-11 and then
started the IP code -- basically RTR style today.   Later it morphed into
Rob's Vax BSD  4.1 specific stack,  which we ran at UCB on a couple of the
systems using 3M Xerox board.  This latest until 4.1A and Joy's rewrite and
I want to say we switched in Interlan 10M boards then.  We have a couple of
the 3Com boards, but because of the lack of buffering, they were a bear to
use and stopped as soon as we got the Interlan one.


Anyway, all of these IP/TCP stacks used Rob's mbuf code.  Which was a
blessing and a curse.  By writing his own, he avoids huge
changes/integration into the memory system, but it also helped to make BSD
such a mess under the covers because there were so many private memory
managers between the network, the I/O systems etc...  As discussed
previously on the TUHS list, the one thing Risner really did well had a
uniform memory design.   Later BSD's moved to Mach and tried to clean this
up a little, but the network code was by then so screwed into Rob's mbuf
scheme, it stayed around a long time.  Werner -- what is the state of this
these days in FreeBSD is it still there?




> There are recent optimizations to help the CPU with prefetch, and some
> ideas around vectors of mbufs.  What's remarkable is the mbuf design
> scales to
> 200gbps in practice, it must feel great to design something like that so
> long ago :)
>
Well, ask Rob :-)  I've lost track of him since Stellar, and I think he I
heard he left high tech but frankly don't know.

Clem
ᐧ