In late 1983, the Multics Cray Station software had just been completed and installed at a site in France (Centre de Calcul Vectoriel pour la Recherche, or CCVR), where earlier that year I had done some consulting work getting the Multics system installed and operational. This was very impressive software: at the time, it made Multics the fastest Cray front-end around, which was pretty impressive considering that the competition included big CDC scientific supercomputers. The Cray station software was written by Warren Johnson (HIS), Michel Senelart (Bull), Guy Autier (Bull), Jackie Barre (INRIA), and John Churchhouse (HIS-UK). It topped off at about 45K lines of code, some of which was ALM for performance and direct hardware control. The Multics hardware connected to the Cray system using a Hyperchannel local network, running at the then-phenomenal speed of 50 megabits/second over a cable somewhat thicker than the average garden hose.
With this software in hand, Honeywell was eager to sell a third Multics system to Ford Motor Company as a front-end for their soon-to-be-installed Cray. The Cray wasn't yet at Ford, but there sure was a lot of Multics hardware, so they came up with a plan to connect over a telephone line to a Cray system at the Cray facility in St. Paul, Minnesota. This would allow a demonstration of the software's capabilities--although definitely not its speed, because the best modems in the world (which we had, more or less) ran only at 9600 baud. Being big Multics fans, Ford was eager to participate, so we got the Hyperchannel boxes in place and set things up.
Bernie Licata, the salesman for the Ford account, wanted to be sure he had a system expert on hand in case anything went wrong, so he hired me to come out to Ford and assist (for "a a few days"... I ended up staying two weeks). Because I hadn't actually had anything to do with the Cray at CCVR (I believe the system didn't even arrive until after I'd returned to the U.S.), I puttered around a bit and watched while Warren set up the Cray station software and we got the Hyperchannel boxes (each the size of a small suitcase) connected. There weren't any "system" issues for me to help with, so I mostly spectated.
At each end, there was a Hyperchannel I/O adapter connected to the computer (either the Multics or the Cray). Each I/O adapter was connected (by that heavy coaxial cable) to a Hyperchannel communications adapter, and that adapter was in turn connected to a modem by some daintier wiring. The idea was that we'd connect all this stuff together, then establish the telephone connection and connect the two modems to it, and everything would "just work". Each side of the lash-up (the Ford Multics in Dearborn, Michigan, and Cray system in St. Paul) had checked out perfectly all the way to the modems, so it seemed like the final step would be easy.
So, we made the telephone connection. The modems recognized each other and showed ready, so we started trying to communicate. Nothing. Data lights would blink, as if traffic was going back and forth, but neither system could see any activity. We tried a bunch of different approaches, but nothing worked: lights were on (and blinking), but nobody was home.
This was very discouraging, but it was pretty clearly not a Multics software problem, so I started reading the Hyperchannel documentation. Eventually, deep in one of the manuals, I discovered that the model of Hyperchannel communication adapters that we had included a firmware limitation that permitted them to talk only to full-duplex modems--and our modems, of course, were half-duplex, as 9600-baud dialup full-duplex modems were well in the future back then.
How had we overlooked this detail? Well, it wasn't entirely clear, but because we'd gotten explicit instructions from the Hyperchannel engineers about how to configure the adapters for half-duplex operation, it didn't seem wholly unreasonable for us to have assumed that they would work in that configuration! Upon inquiry, they allowed as how they'd overlooked that limitation with our specific adapters, and that we couldn't use them after all. We couldn't see any way to get replacement Hyperchannel adapters in time, and everybody started getting nervous: we only had a few days left before the window for our demonstration closed.
At least now we had an explanation, but we didn't have a solution. I started talking it over with Fred Burford, Ford's excellent SiteSA, and we came up with an idea: if we could get another pair of modems, and another phone line, maybe we could pretend that the two pair of half-duplex modems were actually a single full-duplex modem. We told Bernie about this, without getting into much detail, and he swore he'd find us another pair of modems. Fred and I took off for Radio Shack to buy cables, connectors, and time.
We came back and soldered together a pair of three-headed cables designed to split the transmit and receive sides of the synchronous full-duplex interface. Neither of us had ever worked with a V.35 interface before, so we were somewhat unclear about the details, but we did the best we could. While we were doing this, Bernie came back to say he'd found a pair of modems and was having another salesman bring them over. We then explained more about what we were doing, and how we'd been out to Radio Shack to get the parts. Bernie looked cheerful and optimistic as always, but he told me much later than when we'd gotten to the part about Radio Shack, he saw his whole commission flashing before his eyes.
Soon, the new modems arrived. They looked OK, except for one minor hitch: the first pair of modems was 9600 baud, and the new pair was only 4800. At first glance, this seemed like a catastrophe, and we started worrying about how to get yet another pair of 4800-baud modems just so they'd match, but I looked more closely at the V.35 specification and realized that the receive and transmit clocks seemed to be completely independent, so maybe it would work even though one side of the channel was half the speed of the other.
In any case, we had no better ideas, and it was getting late. We gave one modem and one of the special cables to a field engineer and hustled him off to the airport, where he hopped on a plane and headed for St. Paul. When he got there, we talked him through connecting everything together, and then we tried to make the phone calls. After a short comedy of errors in which we repeatedly called a tire dealer in Eagan, Minnesota several times before getting the right number, we got both phone lines connected, both pairs of modems talking, and both Hyperchannels apparently happy. Warren started up the Cray station software, and it worked immediately.
After that, everything went smoothly--the software worked, jobs ran on the Cray, Ford bought the third Multics system, and they got a lot of use out of their Cray. Few things become more thoroughly obsolete than supercomputers, however. I recall being told in the early 1990's that the Cray was no longer used for anything computational--instead, it was serving out its declining days routing print jobs!