[HECnet] Connecting an HP Itanium BL860c to HECNET... without ethernet

[Paul Koning]

> On Dec 10, 2021, at 7:08 PM, Thomas DeBellis <tommytimesharing at gmail.com> wrote:
> 
> Those names almost sound like they want to be familiar, but not quite.  What I recall for a DN20 was that, in terms of communication, MCB only had two speeds.  
> 
> 	• A 'fast' interface could do 56 KBid, synchronous.  I think that part was called a "KMC".  It may have had some sort of embedded microprocessor on it, but I don't recall what kind of operations it performed.  
> 	• A 'slow' interface could do up to 9.6 KBid (I think).  I can't quite remember what it was called.  Maybe a DUP-11?  I might be getting that mixed up with the RSX20F front end, which had single line interfaces that I think were called DL's.

The only system I know well is RSTS/E.  The only sync comm link it supported were DMC11 and its offshoots DMR11 (same features, modernized design), DMP11 (added multipoint) and DMV11 (like DMP except Qbus).  All except DMV could run up to 1 Mb/s and in fact that speed was achievable.  DMV ran at a max of 56 kbps, I'm not sure why.  That board has a very different design, much smaller.  The DMC/DMR/DMP are all variations on the KMC11 microcontroller, with a directly-connected line card.  There actually were two line cards: one "integral modem", coax connected for up to 3 miles or so distance at speeds in the 56k to 1M range.  The other was a modem card (RS232, RS422, V.35) for up to 56 kbps if the modem could go that fast.  (Perhaps more but I don't know of modems that would do so).

The KMC you mentioned was a programmable engine which could do DMA and Unibus transactions, so it was paired with sync or async comm devices to offload DDCMP processing.  The idea is that the host would talk to the KMC with a packet API, and the KMC would then talk on its behalf to the actual UART or USRT with byte-oriented transactions to do the real communication.  You'd still have a whole lot of bus activity but no character-level interrupts to the host.  I don't know what speeds were achievable with KMC-DUP or KMC-DZ, the two pairs I know of.   Part of the question is what the most is those character devices could do.  Someone else might have those details.  RSTS never supported KMC for DECnet, though it did for RJ2780 I think.

DUP alone can also be used, in that case  you'd be looking at DDCMP in software and a driver directly controlling a character device.  That wasn't in RSTS either, but as I mentioned I did implement driver-level DDCMP for serial ports.  That normally meant UARTs, though the comm port on the Pro can do either sync or async and my driver for it supports both modes.  Given the slow CPU and the insanely ugly interrupt architecture of the Pro series, 9600 baud is problematic and anything faster is entirely out of the question.  4800 works fine, and if you have the ability to use odd speeds it's possible to get nice reliable operation at, say, 8000 bps or so.  The reason it's problematic to go fast is that on the Pro you're competing with other high rate interrupts from devices like the console display frame buffer.

	paul