[Simh] IMP interfaces

[Clem Cole]

On Wed, Mar 7, 2018 at 2:30 PM, Phil Budne <phil at ultimate.com> wrote:

> > >Can the IMP software and the H316 simulator talk TCP/IP to an ehernet
> card?
>
> If we had TCP-era IMP code, one could, in theory, but as I recall the
> structure of the TCP ARPAnet (net 10) address was that one octet was
> the IMP number, and another the port on the IMP, which was severly
> limited (four host ports?),

​Exactly...   up to 4 hosts per IMP on the original Honeywell - each
supporting a single 1822 interface, which might or might not be local using
a 'very-distance-host' interface - i.e. was not in the same room as the IMP.

Berkeley's original connection was via the LBL IMP using the VDH​

Noel Chiappa might remember more of the TAC details, I personally never saw
one - although I did use one or two on occaison.   IIRC they could handle
??32?? lines maybe, but there were for terminals (and a lot of modems).
 The whole reason for building the ARPAnet was to allow sharing of
expensive mainframes (of course what it did was actually cause people to
want more systems).   But until the TAC's show up, 'remote sharing' was
less than might have been hoped by the funding masters ;-)

More details of all this is can be found in Katie Hafner's 'Where Wizards
Stay Up Late'




> and if that limitation was "well known"
> ​ ​
> you might only be able to attach to four hosts over Ethernet.
>
​Mumble...  it was certain 'well known' the question is was it screwed down
into SW.   Which is likely.​



>
> Most often, the system attached to the IMP acted as a router, and
> ​ ​
> connected to the local internet.
>
​Actually that's quite right if my memory serves me.​


The original​ idea of the original (Honeywell) IMP was to keep the host
from having to do any protocol work.  The host SW was left as an exercise
for the user by BBN [again see Katie's book - the details are
fascinating].   Until TCP was thought about (which was before IP BTW); the
host was supposed to be very dumb and the IMP terminates the network and do
all the 'hard work.'   I think BBN looked at the 1822 as a way to add a
peripheral to the host in the key of a terminal more than anything else.

In fact the host SW, was not even started until pretty near the time that
the first IMP was delivered.   There was not specs for it for a long time.
That said, as
 time went by and the host SW got more and more sophisticated and
​ eventually​
the need for TCP arose, the IMP was a
​protocol terminator/​
converter more than a router.  With the advent of IP and then running the
​network ​
protocol in the host itself (TCP), BBN created the C30 IMP (20 bit
microcoded UNIX system, they later sold as a UNIX system called the C70).
 That would be more like a router.

I'm drawing on very old memories here...   you'd need to ask some one like
Al Nemeth who managed the project at BBN, but I don't think the C30's ever
ran the NCP code.

In UCB's case when we finally got our own IMP (82/83ish I think) it was a
C30 and that was all post NCP turn off date.    I'm also not sure of the 4
hosts 1822 limit was removed.  I don't remember anyone having used 1822
with them, but I'm fairly sure their had to be because Ethernet interfaces
did not yet exist for many of the original ARPAnet hosts.  My memory is
that the C30's supported native ethernet, so they really did become a
router and people stopped needing to use an 1822 with them.

Clem
ᐧ
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.trailing-edge.com/pipermail/simh/attachments/20180307/052924a3/attachment.html>