dyer@spdcc.COM (Steve Dyer) (02/21/90)
Async support in the original Dec 88 release never worked correctly, and was
essentially unusable in my config (buffered 4 port card, one at 19.2 SLIP,
the other 3 attached to Telebits.) Async support even under heavy input
load works fine in the Sep 88 release.
I'd applied, but never run, the patches to the Dec 88 release (1.3) claimed
to make async support work, mainly because I had a device driver which was
said to be customized for the Sep release. Now that I have a newer version
of that driver, I built a kernel, almost salivating at the prospect of
lower-overhead async support.
The problem is that I'm still seeing input overrun conditions with
this patched async support at a frequency of 10-20 per minute when
incoming network traffic is heavy (I haven't yet seen what TB+ traffic
will be like), compared to no more than one or two per day (if that)
under the Sep 88 release.
This is a great disappointment. What have been other people's experiences?
Should I be resigned to stepping back to the Sep 88 release yet again????
--
Steve Dyer
dyer@ursa-major.spdcc.com aka {ima,harvard,rayssd,linus,m2c}!spdcc!dyer
dyer@arktouros.mit.edu, dyer@hstbme.mit.eduedler@jan.ultra.nyu.edu (Jan Edler) (02/23/90)
Steve Dyer reports that he still has input overrun conditions on his Dec 88 kernel with the buffered 4 port async card, even after applying the 1.3 fix posted here last summer. We don't seem to have any problems, but all we're running is 1 or 2 uucp lines at 9600bps. We did have some problems after applying the 1.3 patch caused by a "bug" in /sys/conf/makefile.ca (missing blank line at end of file) that caused make depend to screw up. The symptom in our case was "Stray Interrupt" messages on the console. When we rebuilt the whole kernel, the problem went away. Async support on the RT has not been one of IBM's strong points (but I'm not sure the competition does any better). The planar ports are almost completely useless under the Dec 88 kernel. There's no real excuse for not having bug-free efficient async support, even on unbuffered ports; I've done it on a much slower machine than the RT. Jan Edler NYU Ultracomputer Research Laboratory edler@nyu.edu (212) 998-3353