Hi,

[I've mentioned this problem earlier somewhere, but never received any
help.]

I have the following problem: when an NFS client connects to NFS server and
starts moving large chunks of data from the server to the client, the NFS
session seems to hang for a long while doing nothing, until it eventually
starts to work. After this, the connection seems to work normally.

In a bit more detail:

  1. I have an alpha running NetBSD-current 1.6ZG. This is the NFS server.

  2. The NFS client is i386 running -current 2.0A.

  3. I boot the i386, and mount NFS shares from the alpha

  4. Let's say I execute the following command from i386:

	# cp /mnt/maailma/netbsd.generic /tmp

     Where /mnt/maailma is NFS share from the alpha.
     This transfer takes about 5 minutes. Observing my switch, the
     lights that indicate network traffic on the i386 and alpha
     interfaces *do not blink*.
     The network is silent, until at the very end it actually moves
     the file.

  5. Let's say I then execute

	# cp /mnt/maailma/usr/share/misc/termcap.db /tmp

     As in (4), this takes forever. /mnt/maailma/usr is different
     fs than /mnt/maailma.

  6. Then, let's say I execute

	# cp /mnt/maailma/usr/share/dict/words /tmp

     The transfer is ready in a few seconds.

  7. Now, if I unmount the /mnt/maailma/usr share and remount it,
     and execute (5) and (6) again, the behaviour is exactly the
     same: first transfer is slow, the ones that follow it are
     normal.

What is the cause of this idleness during the initial transfer?
It of course is not a big problem because it only happens during
the first transfer, but I'd say it is quite odd.

I never experienced this when I had a sparc or slow i386 as the
NFS server...

-- 
Arto Huusko