🚀 go-pugleaf

We are experiencing some severe problems with TCP/IP under DomainOS 10.3.5
on Apollo DN5500, 425s, 425t, and 433s nodes.

Rlogin connections from a node on the token ring to HP 700 and Sun
nodes nodes on our ethernet using an Apollo as a TCP/IP router between
the networks are occasionally dropped.  It appears that the Apollo being
used as a router is sending an ICMP {NET,HOST} UNREACHABLE to to the token
ring only node, which then drops the connection.  We aren't real sure since
we don't have a way to sniff packets on the token ring.  We haven't yet
seen anything suspicious on the ethernet.  The router node shouldn't be
sending these packets, since it is directly attached to both networks.
The other node should not drop an existing connection upon the receipt of
an ICMP UNREACHABLE since this may be a transient condition.

Our ethernet currently carries packets for two IP networks.  The Apollo
nodes on both the ethernet and the token ring see broadcast packets for
the other IP network and forward them to their gateway for this network.
The directed broadcast flag (-b) on tcpd is not enabled.

The TCP/IP software occasionally seems to go semi-catatonic.  If the
node is acting is a router, it continues to broadcast routing updates,
but it can't be pinged.  It may come back to life after a while, or
the node may have to be rebooted.  This problem may only happen on the
token ring side.
-- 
Don "Truck" Lewis              Phone: +1 916 478-8284   Silicon Systems
Internet: gdonl@ssi1.com       FAX:   +1 916 478-8290   138 New Mohawk Road
UUCP: {uunet,tektronix!gvgpsa.gvg.tek.com}!ssigv!gdonl  Nevada City, CA  95959