Ariel@RELAY.PRIME.COM (Robert Ullmann) (10/30/90)
Hi, [ I am copying a larger than usual amount of text from the previous message, and sending to both TCP-IP and TCP-ISDN; I think the discussion can/should continue on TCP-ISDN. To subscribe to TCP-ISDN, send a note (content irrelevent) to TCP-ISDN-Subscribe@List.Prime.COM ] Re RFC1183, ISDN, X.25, and RT records: > From: Drew Daniel Perkins <ddp+@andrew.cmu.edu> > > I also find it confusing exactly how the RT record is intended to work. > > Here's one scenario that I think these might be useful in. Imagine two > isolated island networks NA and NB, such as might exist in two companies > or universities. Each island network has a one or more routers, maybe > RC and RD, with ISDN connections which can establish links to other > networks. Now let's say I try to telnet from one non-ISDN machine HE on > network NA to a non-ISDN machine HF on network NB. One way I could > imagine things working is that router RC might advertise a default route > to hosts on its network including HE. So, when I telnet, HE would send > its packets to RC. Now, when RC received the packet addressed to HF, it > might look up RRs for HF and it might find an RT RR for RD. Looking up > RD it might find an ISDN RR with RD's number. It could then forward the > packet through to RD, which would then forward it to HF. HF's response > would hopefully be sent back through RD (due to having correct updated > routing information) which would then realize that it already has a > connection open to RC, and would simply send it through. For someone who finds it confusing, you have given a perfect example :-) The routers, of course, don't see datagrams as requests/responses, but route in each direction independently. (datagrams from HF in the example route back to HE much as the original path from HE to HF was set up) Usually this will result in the same path in both directions, but it doesn't have to. (no, I don't think that is a problem: given a choice between not reaching a host, and reaching it by an asymmetrical path, I'll take the latter :-) Note that they don't have to be "isolated island networks" for this to be useful: for example, Prime runs its internal net using a private X.25 network, configured as part of several PSDNS; some (most) of Prime's traffic would be inappropriate on the Internet-proper (to, say, one of our customers), being purely commercial. > When we start scaling this simple example up, I see a lot of problems > start popping up. Imagine I just have a single host (at home maybe) > with an ISDN connection. Now, I want to telnet to a host which is > somewhere on the Internet, behind maybe dozens of router hops. It's > hard to imagine that every host on the internet will have an RT RR in > the DNS. If I were a random site with hosts on the Internet, but no > ISDN connections, what RT RR would I want to advertise? There might be > thousands of other hosts/routers on the Internet with ISDN connections. > Which would I choose, and how? > Drew Good example. First, note that your host is going to need an initial default route to bootstrap itself: another host (the one at work, presumably, or perhaps a service like uunet or PSI). It has to know a priori the Internet and ISDN or X.25 address of that host. For hosts that do not have RT or ISDN records, you continue to use the default(s). This may be desired behavior anyway, since reaching connected-internet hosts is probably best accomplished via your work/service. By "connected-internet", I mean hosts on nets known to the core routers. How do they reach you? Two cases. In the first case, you have an IP address on a net advertised to the core (by your service host); they reach you by routing in the normal way, your service host finds you a priori. Second case: you are on a net _not_ advertised to the core. No, they can't reach you if they don't use RT/ISDN themselves. But they couldn't reach you anyway ... (<grin>) About every host (or domain, using wildcards at some level) having RT or X.25/ISDN records: IMHO, an ISDN interface is going to be commonplace soon; sites (even hosts) that do not have their own will be very unusual indeed. (Put Down the Flame-Thrower, Sir! Go back and take note of the "IMHO" :-) IMHO: (<-- note again :-) with the extension of the Internet into more and more commercial domains, and extensive use that would be completely unacceptable to the research Internet (Prime and customers exchanging EDI, for example), the new routing core must be based on the ISDN, with real unrestricted access between organizations. Best Regards, Robert Ullmann Prime Computer, Inc. +1 508 620 2800 ext 1736