🚀 go-pugleaf

In article <4037@utah-cs.UUCP> donn@utah-cs.UUCP (Donn Seeley) writes:
>... flow control and protocol overhead limit the throughput
>to ~300 cps in layers.

The protocol is multiplexed, full-duplex, error-correcting packets
with relatively little overhead for large data transfers.  However,
if you transfer a single byte per packet, the overhead is naturally
substantial.  (The original "jerqio library" implementation suffered
from this.)

Rob Pike reported over 16,000 user bits per second on a 19,200 baud
line, on 8th Edition UNIX.  (He also reports no more than 6,000 baud
effective rate on other implementations.)  If you're using a user-mode
multiplexer based on ptys on a 4BSD kernel, then the context switching
alone will eat you alive.  Doug Kingston published a kernel-mode 4BSD
multiplexer modeled after AT&T's "xt" pseudo-device driver that runs
much faster.

>Another problem with the 5620 has been lack of software.  Rob Pike
>demonstrated features of the Blit software at the Santa Monica Usenix
>back in 1982 which I still haven't seen on a 5620.

Dave Prosser's "myx" package (available rather inexpensively through
the AT&T UNIX System Toolchest) has most of those features.

One point that people who are not heavy DMD users keep missing is that
the DMD is designed to be used for more sophisticated user interaction
interfaces than simple windows.  A typical DMD layer is a downloaded
program that has a fairly capable 32-bit processor available to drive
the user interaction when it needs it.  I have seen some amazing user
interfaces, and I'm expecting to develop yet another in a few weeks.

People who use the default terminal emulator in a DMD layer to run
something like EMACS on the host computer are not exploiting the DMD's
capabilities very well.