*BSD News Article 4956


Return to BSD News archive

Newsgroups: comp.unix.bsd
Path: sserve!manuel!munnari.oz.au!uunet!usc!sdd.hp.com!network.ucsd.edu!qualcom.qualcomm.com!servo.qualcomm.com!karn
From: karn@servo.qualcomm.com (Phil Karn)
Subject: Re: [386BSD]
Message-ID: <1992Sep13.063008.3955@qualcomm.com>
Sender: news@qualcomm.com
Nntp-Posting-Host: servo.qualcomm.com
Organization: Qualcomm, Inc
References: <p5tullo@sgi.sgi.com> <Btr5F5.AJ9@pix.com> <Aug.29.15.45.59.1992.3059@athos.rutgers.edu>
Date: Sun, 13 Sep 1992 06:30:08 GMT
Lines: 21

In article <Aug.29.15.45.59.1992.3059@athos.rutgers.edu> hedrick@athos.rutgers.edu (Charles Hedrick) writes:
>Character loss is normally in the device-level code.  If you're using
>a normal PC serial interface, the kernel must respond to an interrupt
>for each character before the next is ready.  At 9600 bps this
>requires an interrupt latency somewhat less than 1 msec. [...]

I've spent a *lot* of time minimizing interrupt latency in my code.
Even so, I strongly recommend using NS16550As with their 16-byte
FIFOs. With 16550As, even a lowly 10 Mhz 80286 running NOS easily
handles solid streams at 115.2 kb/s from a V.fast modem. The FIFO high
water mark immediately goes to 13 (the trigger point is at 4) so it's
obvious that the port would roll over and die without the '550.  Given
the greater complexity and context switching overhead of the 386BSD
kernel, I'd expect it to be much less able to keep up with a fast SLIP
link. (This is not meant as a slap at 386BSD, since that'd be
comparing apples and oranges. Or more precisely, timesharing systems
and dedicated routers. But sometimes there are significant advantages
to using dedicated routers.)

Phil