dplatt@coherent.com (Dave Platt) (10/11/88)
We, too, had noticed that our Macs tended to "lock up" occasionally
when accessing AppleShare volumes mounted from an Aufs file-server on
our Sun 3/280. The AppleShare activity "arrows" on the menu bar would
appear, and remain on solidly for up to 30 seconds; the Mac could not
be interrupted during these lockups.
I suspected that the problem might occurring if our K-box were to lose
packets due to buffer congestion, and if Aufs were tardy in
retransmitting the lost packets. I checked the Aufs code, and found
that Aufs appears to be setting an entirely-reasonable retry timeout (4
seconds, I believe). Hence, the loss of a single packet could account
for a short delay in AppleShare file I/O, but not a 30-second delay.
A 30-second outage would seem to indicate that the original packet, and
several retry packets were all being lost.
I tried adding the "-S 1 -R 1" options to our Aufs daemon's
command-line, in order to reduce the number of packets sent during each
transaction. This had little or no effect on the problem.
I then reconfigured our "papif" print filter/driver to make the same
sort of change: I reduced the size of the send queue from 8 packets to
1, and reinstalled papif.
Voila! The hung-AppleShare-I/O problem was greatly reduced, and may
indeed have been eliminated. I've seen some delays, but none exceeding
about 5 seconds... it's as though a packet is being lost occasionally,
but the first retry is getting through OK. We haven't noticed any
significant decrease in printer throughput.
Conclusion: it may be possible for a single "papif" stream running
"flat out" (8 packets per burst, with fast turnaround from the printer)
to swamp the buffers in a KFPS-3 running KIP 06/88. Even if the
printer connection doesn't seem to be suffering from lost packets,
other connections through the same K-box may suffer from decreased
performance. It's possible that other packet-intensive applications
(e.g. TELNET running FTP transfers with a large TCP window) could have
a similar effect.
I'd be very interested to hear whether people have a similar problem with
a stock KFPS-3, and whether the same remedy works for them.
It's possible that this problem could be solved more efficiently by
upgrading from a KFPS-3 running KIP to a KFPS-3U running K-Star. The
added memory in the -3U might make the difference between lost packets
and successfully-received packets.
--
Dave Platt VOICE: (415) 493-8805
USNAIL: Coherent Thought Inc. 3350 West Bayshore #205 Palo Alto CA 94303
UUCP: ...!{ames,sun,uunet}!coherent!dplatt DOMAIN: dplatt@coherent.com
INTERNET: coherent!dplatt@ames.arpa, ...@sun.com, ...@uunet.uu.net