MP VAXen, was Re: PDP-8/e and terminals in West Sussex, UK

DEC deliberately made it difficult to get a KA630 without a MicroVAX around it - as it was explained to me, they didn't want people second-sourcing KA630-based machines. The KA620, which was just a KA630 diddled slightly so that VMS wouldn't run on it, was much more available. (The difference is, P0 and P1 page tables live in physical space rather than kernel virtual space.) But if you _did_ have a spare KA630, you could slap it into any AB/CD slot in a MicroVAX chassis, plug in the ID module and set it for an ID other than zero, and - assuming software - run a dual-CPU MicroVAX. I know about this because, in my time in academia, a researcher at the lab I was working at did just this for robot control - one CPU for the main OS, one to run the control law. (I wrote the (tiny) kernel that ran on the second CPU.) He actually found a hardware bug in the KA620/KA630 IPC hardware - it dropped some small percentage of interprocessor doorbell interrupts. Once located, DEC said it should never have worked at all; the only reason it did was that the board was overdesigned by a factor of something like five or ten - an etch run crossed the whole board and, when combined with the pullup resistor, the signal rise time was usually but not quite always quick enough. The fix was a reduced pullup resistor. I think that, as long as ECOs for KA6[23]0s were available, you could get the fix done for you just by knowing what to ask for; they didn't proactively push it out because most people weren't doing multiprocessor machines, and most of the people who were didn't mind losing a tiny fraction of the doorbell interrupts. (I think it was something on the order of single digits per thousand....) /~\ The ASCII Mouse \ / Ribbon Campaign X Against HTML mouse at rodents-montreal.org / \ Email! 7D C8 61 52 5D E7 2D 39 4E F1 31 3E E8 B3 27 4B