On 2/28/2020 9:11 AM, Clem Cole wrote:

[1] The original PC/AT used the NS8250 UART with no input buffering, which went through a couple of generations, eventually begat the *550 version and had I think an 8 character input buffer.  But IIRC none of them had hardware flow control.   I forget the # now, Moto made a nice dual UART with 16 chars of input buffering, that many of us on Unix workstation business used, but when we moved to BSD 386 and Linux, we were stuck with PC hardware, which had a particularly hard time with things like the Trailblazer  (which was the modem of choice for UUCP).

I ran a BBS for a few years back in the early 90's, and used a 486DX2-66 as my "front-end" to a Sun SPARC-IPC USENET setup. Using two V.34 and one Worldblazer, running them at 38,400 baud, and taking advantage of compression, it ran 100% download, 100% upload, or a combination across three modems without even showing much load at all. It could have easily taken more if I had the physical space (and the IRQs) on the ISA bus to add more serial ports. Of course, the interrupt coalescing of the 16550's helped a lot. And I don't know what the saturation point was...

That was on x86 SVR4.2 (Consensys), using a shareware 16550 driver of the time. The Worldblazer talked to a Trailblazer at Motorola for my USENET feed and used G protocol, acceleration built into the Telebits.

8250's and 16540's were horrible. Much like DZ11's, eh? ;)

art k.