Subject: Re: CAP under NetBSD/mac68k
To: A. Forest Godfrey <ag5c+@andrew.cmu.edu>
From: David Hornsby <djh@cs.mu.OZ.AU>
List: port-mac68k
Date: 04/07/1996 22:01:16
> From: "A. Forest Godfrey" <ag5c+@andrew.cmu.edu>
> 1) atlook gags on the Carnegie Mellon campus network. Sometimes it
> core dumps, other times it complains about an invalid name length, and
> sometimes it does both. It NEVER displays all the nodes on the network.
In cap60/samples/atlook.c, change the line
#define NUMNBPENTRY 250 /* max names we can lookup */
to a bigger number.
> 2) A bigger problem is that the AppleShare server (aufs) is
> EXTREMELY slow! So slow as to make it unusable. It took 5 minutes to
> write a 12k file yesterday on a machine that is on the same hub as the
> server!
As an aside, AppleTalk/AFP is very unforgiving of even minor packet losses,
Packet Loss Time-LT Throughput-LT Time-ET Throughput-ET
-------------------------------------------------------------
0% 200 s 100% 15 s 100%
0.01% 205 s 97% 20 s 75%
0.05% 224 s 89% 39 s 38%
0.1% 248 s 81% 63 s 24%
0.5% 424 s 47% 255 s 6%
(LT = LocalTalk, ET = EtherTalk, Time is for 1000 packets, numbers by
Tom Evans).
The usual suspect on a BSD host is having the Berkeley Packet Filter
return multiple packets in a single read. CAP copes with this OK assuming
that MULTI_BPF_PKT gets defined in cap60/support/ethertalk/bpfiltp.c
On NetBSD this is handled by #ifdef __NetBSD__
Check that your sources have this in place.
Otherwise check that your ethernet card is working OK by using ping or
spray etc. to send a bunch of packets to the CAP host. Check for log
or console error messages.
- David.