jeff@satchmo (Jeff Detterman) (02/01/91)
Has any one had any luck connecting apollo token rings via Cisco router boxes. Due to the merger of two of our divisions we need to connect two distinct rings together. The rings are connected via ethernet, separated by a Cisco router. An individual at our company says that this is not possible but I do not believe him. We would like to run DDS services between the rings and eventually merge all services (glbd, rgyd, etc ..). Also is it possible to have two rings connected via ethernet and token ring back each other up. In otherwords how do I make the ethernet take over if the token ring fails. I have assigned Apollo network addresses to the ethernet and have enabled Internet routing on the ethernet gateways. The rings are connected via a DN4500 with two token ring cards and via two DN4500 that have both ethernet and token ring cards in them. Any comments or suggestions would be a great help. Thanks in advance. -- Jeff Detterman, Rockwell International, Cedar Rapids, Ia Email : j_detter@hwking.catd.cr.rok.com
csfst1@unix.cis.pitt.edu (Charles S. Fuller) (02/01/91)
We've been running DDS over cisco routers for some time now, using both dedicated lines and switched virtual circuits, with minimal problem. There are a few minor things that you'll run into, but the overall implementation works quite well. It has been our experience, though, that IP data transfers through the same router perform better than DDS transfers (eg., ftp vs /com/cpf). The first thing you'll notice when configuring the cisco is that they don't provide a node id; you'll have to make one up. We chose numbers that are about 3X higher than current Apollo node id's, to avoid collisions for a little while :-) Enabling routing is a simple task, but you'll have to remember when reading the docs that cisco manual writers have probably never seen an Apollo, let alone configured one. Between myself and a datacomm guru, we were able to arrive at common terminology. Then, once packets start flying, you'll probably want to do an 'lcnet -full -hw -conn' -- just to see what you've done. The first thing you'll see is an error message to the effect that the node identified as the cisco did not respond, and that "touching" information may be incomplete. This is not an indication of problems, but, rather, an indication that cisco routers lack the ability to respond to "higher-level" DDS queries (like 'lcnet'). When combining networks, be sure to give some thought to administrative issues, most importantly the registry. There are 7 different registries in our network, some of which span multiple sites. Some are SR10, some SR9.7. We found out early on that, despite documentation to the contrary, it is not necessary to merge registries UNLESS everyone needs access to every machine. If your sites are to remain autonomous, you may want to leave the registries separate. This also provides a little extra protection from users at other departments, although root is always root, capable of walking through the entire network. Keeping registries separate became a little difficult at SR10.0 and SR10.1, but the addition of NCS cells at SR10.2 made registries a little more manageable once again. Hope these ramblings are of some use. Chuck
jab0396@cec1.wustl.edu (John A. Breen) (02/03/91)
In article <86164@unix.cis.pitt.edu> fuller@nye.nscee.edu (Charles S. Fuller) writes: >The first thing you'll notice when configuring the cisco is that they >don't provide a node id; you'll have to make one up. It's been a while since we hooked up to the ethernet (and I had retired from sys-adminning just before that, so I just watched), but as I recall, the ciscos were pretty much transparent (they were already set up for general ethernet connections). I think we just assigned network IDs to each ring and one to the ethernet as a whole, and that was it. In fact, I remember when we enabled routing, we saw a network we didn't recognize: it turned out to be a ring at one of our facilities on the west coast (although that did cause a minor problem, since they had assigned a different network ID to the ethernet). But from the Apollo filesystem point of view we've had virtually no problems. Now, subnetting for TCP/IP is a different problem; some day I'll have to relate that horror story... John A. Breen McDonnell Douglas Missile Systems Co. ----------------------------------------------------------------------------- John A. Breen | "Where is fancy bread, jab0396@cec1.wustl.edu | in the heart or in the head?" johnb@hobbes.mdc.com | -- Willy Wonka
mort@apollo.HP.COM (Stephen Moriarty) (02/07/91)
In article <1991Feb1.000434.244@zoot.avgrp.cr.rok.com> jeff@satchmo (Jeff Detterman) writes: >Also is it possible to have two rings connected via ethernet >and token ring back each other up. In otherwords how do I make >the ethernet take over if the token ring fails. A lot of people ask for this. You can't do it. The reason is that the registries/naming servers are fixed, and can only be changed explicitly. What you're asking for is that the naming server be dynamic, trying different net.nodeid pairs upon failing to access on. The `node_data/hint_file comes into play, as well, since that's where you get your net.nodeid pair after accessing a node for the first time. If you consult Chapter 7 of, "Managing Domain/OS and Domain Routing in an Internet" for "How to move a node from one network ID to another", you'll see what needs to be done dynamically in more detail. mort ARPA: mort@apollo.hp.com UUCP: ...{decvax, umix, mit-eddie}!apollo!mort Apollo, a subsidiary of Hewlett Packard, 300 Apollo Drive, Chelmsford, MA. 01824 Argue for your limitations, and they are yours.