Path: utzoo!utgpu!jarvis.csri.toronto.edu!mailrus!tut.cis.ohio-state.edu!ucbvax!pasteur!helios.ee.lbl.gov!ncis.tis.llnl.gov!lll-winken!scooter!neoucom!wtm
From: wtm@neoucom.UUCP (Bill Mayhew)
Newsgroups: comp.sys.next
Subject: Re: >= 9600 baud on the serial ports?
Summary: Kernel context switchig becuase of interrupts on 8530?
Keywords: Trailblazer, Fast Terminal
Message-ID: <1595@neoucom.UUCP>
Date: 24 Apr 89 17:02:16 GMT
References: <275@nueces.UUCP>
Organization: Northeastern Ohio Universities College of Medicine
Lines: 32


I find it somewhat surprising that the Next has problems at 9600
baud and/or above.  It is my understanding that the serial I/O for
the two ports on the back is handled by a Zilog 8530 (or second
sourced by AMD I think).  That chip should be perfectly capable of
dealing with it.

Sounds like the driver needs some work.  Maybe they are using an
interrupt per character mode.  Such programming would make a couple
of context switches per character received.  I've seen drivers on
other Unix systems with similar problems, Sun systems for instance.
I don't know too much about the Next's heart of mach, but I was
under the impression that mach's forte is supposed to be dealing
with quasi real-time environments.

Hopefully, this will be fixed by the time 1.0 hits the streets
indeed.  There is no excuse for the Next not to be able to cut the
mustard with the serial port.  My Unix PC (10 MHz 68010) has an
8274 serial controller, which is similar to the 8530, and my Unix
PC can achieve a sustained throughput of ~1400 char/sec when
running uucp over a trailblazer modem.   The 8274 is desinged to
optionally implement DMA, but must be used with an external address
generator; I don't think that was done in the Unix PC.

By the way, can the Macintosh receive a steady stream of data at 56
kb/s?  It is prettty hard to do that without a 2-port ram buffer or
a DMA channel from the serial port .. or else shut the system down
to all other operations while receiving.  At that rate you start
running up against interrupt latency problems if you don't.

Bill
wtm@impulse.UUCP