Peter Smithson wrote:

> I've just remembered another question.  The program takes about 18 hours
>to run so I modified it to display it's progress to the screen.  I
>remember reading something about how using screen IO can really slow
>down a process on MPE.  Is that right?
>
>It only displays a message roughly once every 40 seconds.  Is that going
>to slow it down much? I guess I can run again without the DISPLAY line
>but it takes so long to run.
>
Well, it does take "some" cycles to display, but since MPE/XL (RISC
family) terminal I/O doesn't really matter very much, if you're talking
about a real terminal connected over a serial line.  In the days of
classic MPE, there was a considerable timing issue since every character
generated an interrupt and the system had to process it on the ICS,
slowing down everybody.  PA-RISC solved this with the DTC, which handles
all the individual character interrupts and buffers the I/O to the system.

Having said that, there are now network connections to the system.  The
old original NS/VT protocol acted much like the DTC, buffering
characters (in the client) and interrupting the system only when a
complete line or I/O was performed.  Next came the telnet protocol,
which was a gigantic step backward initially.  Raw telnet also requires
interrupt-driven events on each individual character.

Most recently, and Wirt is probably the best spokesperson for it, is the
"advanced telnet" support as implemented in QCTerm that uses an
extension of the original telnet specification (not supported by much of
anything other than QCTerm that I know of).  This mode utilizes
half-duplex input and buffers I/Os in the client not unlike NS/VT.

With all that said, the overhead is mostly dealing with INPUT.  There is
not much difference in any case for OUTPUT, as long as you aren't trying
to output faster than the connection in question can support.  So one
line every 40 secs is trivial.

Jeff

* To join/leave the list, search archives, change list settings, *
* etc., please visit http://raven.utc.edu/archives/hp3000-l.html *