Linux Virtual Server and tulips

[John Connett]

I've been experimenting with the Linux Virtual Server:

    http://proxy.iinchina.net/~wensong/ippfvs/

and have encountered a problem when using tulips (both the old style
Kingston KNE 100TX (21140-AF) and the new style (21143-PD) which occurs
when the Network Address Translation (NAT) and LocalNnode options are
specified.  Both the v0.90 and the v0.91 drivers exhibit the problem. 
This occurs with the latest vs-0.9 patch and also occurred with the
vs-0.8 patch.  The symptoms are that a virtual server connection
redirected to the local node times out without establishing a
connection.  Connections to slave server nodes work as expected.

When I replaced the tulip with an Intel EtherExpress PRO/100+
(eepro100.c:v0.99B 4/7/98) the problem does not occur.  Wensong Zhang
has reported that the problem does not occur with an NE2000 NIC.  Are
there any obvious ways in which the tulip behaviour is different from
these other two cards?

Any hints as to how I might track down this problem and find a fix would
be most gratefully received.  Unfortunately the hardware is fixed for my
particular application so replacing the tulip is not an option!  The
final target is the 2.0.36 kernel on Alpha but development testing has
been on a Pentium III.

Thanks in anticipation
--
John Connett (jrc@art-render.com)