Path: utzoo!attcan!uunet!lll-winken!lll-tis!ames!ucsd!rutgers!bellcore!tness7!texbell!bigtex!james
From: james@bigtex.uucp (James Van Artsdalen)
Newsgroups: comp.sys.ibm.pc
Subject: Re: ISDN
Message-ID: <7340@bigtex.uucp>
Date: 31 Aug 88 22:48:33 GMT
References: <23056a96@ralf>
Organization: F.B.N. Software, Austin TX
Lines: 41

In article <23056a96@ralf>, Ralf.Brown@B.GP.CS.CMU.EDU writes:
> In article <6108@bigtex.uucp>, I wrote:
| [...] Mess-DOS
| rolls over dead at the thought of 9600bps (interrupt latency on
| extended memory on slow machines), much less 8Kbytes/sec!  I rather

> Better not tell that to all the 4.77 MHz IBM PC's running USRobotics
> HSTs at 9600 or even 19200....

Hmm...  It's entirely possible that the problem is strictly related to
extended memory.  Extended memory on ATs leaves interrupts turned off
for a very long period of time, and this is factor if you're using a
disk cache, as most people seem to these days.

| [...]  Even using an NS16550A serial chip with the 16 byte FIFO
| isn't going to cut

> 6400 / 16 = 400 interrupts per second.  Sounds quite manageable to me,
> even on a 4.77 MHz 8088 box.  When I still had my Eagle PC (4.77
> 8088), I used to run of CPU cycles.  [...] If a serial port handler can't
> hack 1/12 the interrupt rate, something is wrong....

The problem isn't the serial port, or even the interrupt rate, but MS-DOS
and the design of extended memory systems.

| [...] not that the Mess-DOS file system is
| currently good for sustained 8K/sec anyway.

> Even if DOS is quite sloppy and needs two rotations to write a track on a
> floppy disk, it can write two tracks (9K) in 0.8 seconds, which sure looks
> like more than 8K/sec to me....

It might write that if you did a write() with a 9K buffer, but not if
you did 18 writes of 512 bytes, as is the norm for most programs
(standard I/O package influence).  There is a lot of overhead between
a program and the routines that actually write a sector to the disk.
I would not be surprised if 8K/sec were near the limits of what a hard
disk could sustain (assuming substantial interrupt overhead merely
getting the 8K/sec into RAM).
-- 
James R. Van Artsdalen    ...!uunet!utastro!bigtex!james     "Live Free or Die"
Home: 512-346-2444 Work: 328-0282; 110 Wild Basin Rd. Ste #230, Austin TX 78746