steve@calvin.vf.ge.com (Steve Cunningham) (05/11/91)
May 9, 1991
General Electric Aerospace network
1000 Macintosh, 40 routers ( ethernet to local talk)
I am involved in the effort to break this large bridged network
up into 11 sub-nets. We have cisco routers on order (11 IGS and
15 AGS+). We are conducting a demo with Cisco's help and we
learned quite a bit about Appletalk Phase II and Cisco routing.
We quickly learned of the two well known problems :
MCI boards don't talk to each other unless fast caching off
note: this is fixed in a later beta firmware
Random, somewhat transient AARP failures casusing some zones to
float in and out.
note : also fixed in later beta ware
But last week we stumbled onto a new one which was a surprise to
all, but even now, Dave Edwards has built a fix and distributed
(via internet).
The problem is that a cisco ethernet port can confuse itself with
a node which has the same node address, although a different
network or cable address.
eg: cisco port E0 has address 4.
a node ( we had three) had address 4656.48 also legal.
The cisco acted as if he were that node. Symptom, any node with a
xxxx.48 could not see anything on the other side of the AGS+ any
port. Since all of our apples need to acess a mainframe on port
1 these nodes were "broken"
After some furious protocol analysis we realized that packets
(NBP) were passed from the node through the cisco heading for the
named service. Fine. The return packets with address 4656.48
would go into the cisco and not come out. The cisco thought it
was that node and did not forward the packet. We had three nodes
with .48 addresses. By zapping P-RAM we fixed one machine by
making him take a non .48 address. Strong evidence.
Then, we wrote a quick little code whichallowed the reading and
writing of that entry in P-RAM where the
hint for the address is stored. Re boot the Mac and you have
created a xxxx.48 machine. It then showed as "broken". Cause
and effect !
Last thursday 2 May we reported this to cisco. New code ready on
Friday which we downloaded. Booted the AGS+ via tftp and problem
solved. All .48 machines returned to service.
Conversations with Dave Edwards and Patrick Jones assure us that
this fix will be included in the 8.3.(4) release which is due out
officially in one or two weeks.
Hope this adventure is of use to others; for more info I would be
happy to talk.
Steve Cunningham
215-354-4906
General Electric
P.O. Box 8555 blg 19
Philadelphia, PA 19101
Looking forward to sharing other cisco adventures.
steve