History | People | Library | Sites | About Search

Olin Sibert

The Hyperchannel network was an early local-area network, which could run at the then-phenomenal (late 1970's, remember) speed of 50 megabits/second using really heavy-duty coax cables. A Hyperchannel adapter consisted of a generic part that connected to the coax cables and a hardware-specific part that provided an interface to whatever I/O channel hardware was available for various vendors. There were Hyperchannel adapters for Crays, CDC machines, DEC-10s, IBM 370s, and many others, including the Honeywell 6000-series. All this hardware was custom-designed by Network Systems Corporation (makers of the Hyperchannel). I think a Hyperchannel adapter typically cost in the neighborhood of $50K, and you needed at least two of them to make a network.

On Multics, the adapter connected to an ordinary IOM channel. As far as I know, the only two pieces of software ever to talk successfully to the Hyperchannel (aside from the odd diagnostic and such) were the Cray front-end and the TCP/IP implementation, and they both did it through ioi_ (which was pretty straightforward, all things considered).

My very first consulting job (in 1982) was to have produced a third piece of software: a ring-zero multiplexor that would use the Hyperchannel to talk to Modcomp minicomputers. My client (ASEA) felt they needed this because they planned to host a "software factory" for Modcomps on their brand new Multics system, and obviously the two sides needed to talk to each other. ASEA was no stranger to heavy-duty software engineering: they had previously written two whole distinct operating systems for the Modcomp hardware and implemented the "software factory" on previous mainframes.

Why Hyperchannel? Because there was no other interface in common between Multics and Modcomp (except for RS-232 links that seemed sure to be too slow) and for some reason, Network Systems had built an adapter for those underpowered 16-bit machines... never mind that the adapter boxes were nearly as big and expensive as the Modcomps they connected to.

It was an interesting project, and I wrote a fair amount of code for it, but eventually ASEA came to their senses and realized how silly the idea was--as I recall, they gave up on both Modcomps and Multics both around then, and switched to PDP-11s and Vaxes with self-hosted development--no more factory, and no more Hyperchannel. Probably saved them millions of dollars.

In the waning days of CISL, I recall exploring the idea of mounting hierarchies and taking page faults over the Hyperchannel network, but no code was ever written. I imagine this idea was inspired by NFS and/or Apollo. We also talked about using Hyperchannel as a gateway to Ethernet, but I don't think that ever happened either, even though Network Systems undoubtedly produced the necessary hardware. The idea of a $100,000 NIC card was just too much.