It's been reported to me that when an NC tries to boot from a RiscPC
over NFS, this can be unreliable if a Atomwide Ether3 is used as the
network card. The symptoms are a string of errors in the console: 'nfsd
send error 55' and 'dhcp: sendpkt: No buffer space available'. The
problems go away if an Atomwide EtherB is used. This is odd seeing that 
the two drivers are virtually identical. I personally can't recreate
the problem, but then I'm using old Ether3s whereas the installer that
reported the problem in the first place is using brand new ones. The
controller chip is different on the new ones; it's now the same part as 
the EtherB (SEEQ80C04) whereas on all the older cards it is the
SEEQ8005. There's probably not a lot of difference though.

Now, you may remember an earlier posting I made about mbufs and how if a
RiscPC was used as a proxy by more than about 5 clients, the connection
slowed down and No buffer space available was returned when pinging. The
machine in question had client machines on an internal network from an
EtherB (and thus worked fine for anything purely internal) and then had an
ISDN router attached to a new Ether3. Only when there was traffic over the
router due to clients using the server as a http proxy did the slow down
and errors occur. 

The facts seem to suggest to me that the current ea driver has problems
with mbufs when talking to SEEQ80C04 hardware. systat mbufs doesn't
suggest anything. If two network cards are required and new Ether3s don't
cut it, what are the alternatives? I-cubed EtherLan500s are not
recognised (or rather they give the error not configured).

Stephen Borrill