[comp.unix.aux] Long pause using Berkeley networking

dave@onfcanim.UUCP (Dave Martindale) (09/07/88)

We have a few new A/UX systems, configured for Berkeley networking and
NFS but not Yellow Pages.

Whenever I run a Berkeley-style networking command (rcp, remsh), there
is a delay of several seconds before anything happens.  Looking at the
Ethernet wire itself, there is an initial short packet sent out, with
no reply.  Four seconds later, there is suddenly a flurry of packets
exchanged and the results appear.

I don't have an Ethernet monitor, so I know when packets are sent but
not what is in them.  However, it does appear that the Mac sends the
first packet, which is ignored by everybody.  Then, 4 seconds later (by
the stopwatch), it sends another packet, and at that point the remote
machine responds.

Does anybody know what is going on here?  Is the Mac looking for a
Yellow Pages server to look up the remote host name, even though there
is no "+" line in /etc/hosts?

There is no such delay when accessing a file via NFS, so IP and the
Ethernet hardware are not to blame.  However, it does take a very long
time to mount and unmount NFS filesystems (about 4.5 seconds each),
causing me to further suspect something in the host lookup process.

The problem is definitely on the Mac somewhere - the delay is there
running any remote command to or from the Mac, but not between any
other combination of our machines.

Does anyone else run TCP and NFS without YP?

The other machines are:
	VAX 11/780,     4.3BSD
	IRIS 2400T,     SGI 3.6 unix
	IRIS CS16       SGI 4D1-3.0


Dave Martindale
{mcgill-vision,watmath}!onfcanim!dave

liam@cs.qmc.ac.uk (William Roberts) (09/10/88)

In article <16077@onfcanim.UUCP> dave@onfcanim.UUCP (Dave Martindale) writes:
>Does anybody know what is going on here?  Is the Mac looking for a
>Yellow Pages server to look up the remote host name, even though there
>is no "+" line in /etc/hosts?

We are all suffering from the blessed name daemon. The
gethostent(3N) routines are trying to locate a nameserver by
broadcasting on the network to do host lookup. This is what our
monitoring system shows:

  (ethal) trace 10 occurrences of mymc2
  packet filter-name time  etdst               ipsrc usport udport
  0000   mymc2       0000  0x02.07.01.00.34.da aux0  1023   2049
  0001   mymc2       0220  0x02.07.01.00.34.da aux0  1023   2049
  0002   mymc2       0380  0x02.07.01.00.34.da aux0  1023   2049
  0003   mymc2       0600  0xff.ff.ff.ff.ff.ff aux0  1086   domain
  0004   mymc2       37780 0x02.07.01.00.34.da aux0  1023   2049
  0005   mymc2       37900 0x02.07.01.00.34.da aux0  1023   2049
  0006   mymc2       37980 0x02.07.01.00.34.da aux0  1023   2049
  0007   mymc2       38180 0xff.ff.ff.ff.ff.ff aux0  1089   domain
  0008   mymc2       52940 0x02.07.01.00.34.da aux0  1023   2049
  0009   mymc2       53260 0xff.ff.ff.ff.ff.ff aux0  1092   domain

Ok Apple - how do I turn this off? I too would like to mount
NFS filesystems without hanging.

>Does anyone else run TCP and NFS without YP?

Yes, we do. The only people who get NFS are our undergraduate
students and that's only because there machines will die if the
NFS servers go down: a little YP poison can't makes things much
worse. Otherwise I avoid it like the plague because its retry
strategy is so awful: I don't believe that it should try
forever to find a server.

-- 

William Roberts         ARPA: liam@cs.qmc.ac.uk  (gw: cs.ucl.edu)
Queen Mary College      UUCP: liam@qmc-cs.UUCP
LONDON, UK              Tel:  01-975 5250