|
Note 4
FROM: Ben Barker
We certainly did connect the IMPs' host interfaces together to test
them. We tested IMP1 against IMP0, the non-ruggedized prototype. We
also tested the host interface against a host interface tester box that
Marty Thrope designed which set a virtual word length at whatever we
wanted so that we could test the padding circuitry against every
possible word length we might come up against. I wrote the diagnostic
for the 516 to test all this -- it was the diagnostic that I used in
getting the hardware to work.
The story of getting the hardware to work is an interesting one. In
generating the proposal to ARPA, Severo had drawn conceptual block
diagrams of how the interfaces (host, modem, Real Time Clock, Task
Interrupt, Watchdog Timer, etc.) were to work. When BBN won the
contract, these diagrams were simply turned over to Honeywell to design
and build. Honeywell simply implemented the block diagrams in their
logic without ever understanding what they were supposed to do or how.
Naturally, they never had a way to test them or even a concept that they
should. What arrived at BBN in IMP0 was a literal transcription of
Severo's block diagrams wired into logic. Not surprisingly, it was very
far from anything that could work. My challenge in the weeks between
the arrival of IMP0 and the delivery to UCLA was to change Honeywell's
implementations into something that would work -- and to do so in time
to give the software guys some time to debug their code prior to
shipment. The documentation on the interfaces that Honeywell provided
with the machine was simply drawings of the gates and other logic
elements that they implemented.
Truett flew out home to LA prior to the IMP shipping. Frank wanted me
to go on the cargo plane that the IMP went on, but the carrier said it
was not allowed. I stayed with the IMP as long as they'd let me, then
caught the first flight out the next morning. When I got to Bolter,
Truett had the machine unpacked. We powered it up and it came up still
running the diagnostic it had been running before it was shipped. It
ran with no errors. We loaded the operational program paper tape. Mike
Wingfield plugged in the cable to his Sigma 7 interface and started his
host-side diagnostic. This was the first communication between an IMP
and a real (non-test) host. The diagnostic ran but had periodic errors,
every few hours as I recall. I strongly suspected a synchronizer error,
a suspicion reinforced when Mike said that he didn't believe that a
flip-flop could be left in an ambiguous state. We looked at his
drawings, quickly found the place the synchronizer needed to be, and he
quickly wired it up. He restarted his diagnostic and it ran flawlessly.
I called Frank to tell him I was done and ready to come home. He told
me to stay in LA for a few days in case something went wrong. Truett
and I spent a few days bumming around LA while nothing went wrong. He
introduced me to some fine pizza/beer/pool halls. A fine time was had by all waiting for nothing to happen.
The first communication between IMPs at a distance came a month later
when Marty installed IMP2 at SRI. I was at UCLA again. He brought up
his machine with the operational program and I saw the front panel light
on my machine go out indicating that the line was up. I crosspatched
from IMP1 TTY fake host to IMP2 TTY fake host and said hello -- the
first communication (other than IMP to IMP) across the net.
Later, after SRI got their host connected to the IMP, the UCLA guys
tried to log in host-to-host.
/Ben
We certainly did connect the IMPs' host interfaces together to test
them. We tested IMP1 against IMP0, the non-ruggedized prototype. We
also tested the host interface against a host interface tester box that
Marty Thrope designed which set a virtual word length at whatever we
wanted so that we could test the padding circuitry against every
possible word length we might come up against. I wrote the diagnostic
for the 516 to test all this -- it was the diagnostic that I used in
getting the hardware to work.
The story of getting the hardware to work is an interesting one. In
generating the proposal to ARPA, Severo had drawn conceptual block
diagrams of how the interfaces (host, modem, Real Time Clock, Task
Interrupt, Watchdog Timer, etc.) were to work. When BBN won the
contract, these diagrams were simply turned over to Honeywell to design
and build. Honeywell simply implemented the block diagrams in their
logic without ever understanding what they were supposed to do or how.
Naturally, they never had a way to test them or even a concept that they
should. What arrived at BBN in IMP0 was a literal transcription of
Severo's block diagrams wired into logic. Not surprisingly, it was very
far from anything that could work. My challenge in the weeks between
the arrival of IMP0 and the delivery to UCLA was to change Honeywell's
implementations into something that would work -- and to do so in time
to give the software guys some time to debug their code prior to
shipment. The documentation on the interfaces that Honeywell provided
with the machine was simply drawings of the gates and other logic
elements that they implemented.
Truett flew out home to LA prior to the IMP shipping. Frank wanted me
to go on the cargo plane that the IMP went on, but the carrier said it
was not allowed. I stayed with the IMP as long as they'd let me, then
caught the first flight out the next morning. When I got to Bolter,
Truett had the machine unpacked. We powered it up and it came up still
running the diagnostic it had been running before it was shipped. It
ran with no errors. We loaded the operational program paper tape. Mike
Wingfield plugged in the cable to his Sigma 7 interface and started his
host-side diagnostic. This was the first communication between an IMP
and a real (non-test) host. The diagnostic ran but had periodic errors,
every few hours as I recall. I strongly suspected a synchronizer error,
a suspicion reinforced when Mike said that he didn't believe that a
flip-flop could be left in an ambiguous state. We looked at his
drawings, quickly found the place the synchronizer needed to be, and he
quickly wired it up. He restarted his diagnostic and it ran flawlessly.
I called Frank to tell him I was done and ready to come home. He told
me to stay in LA for a few days in case something went wrong. Truett
and I spent a few days bumming around LA while nothing went wrong. He
introduced me to some fine pizza/beer/pool halls. A fine time was had by all waiting for nothing to happen.
The first communication between IMPs at a distance came a month later
when Marty installed IMP2 at SRI. I was at UCLA again. He brought up
his machine with the operational program and I saw the front panel light
on my machine go out indicating that the line was up. I crosspatched
from IMP1 TTY fake host to IMP2 TTY fake host and said hello -- the
first communication (other than IMP to IMP) across the net.
Later, after SRI got their host connected to the IMP, the UCLA guys
tried to log in host-to-host.
/Ben
Copyright 2012-2022 @ xbbn.org under a Creative Commons license.
