Multi-protocol wide area networks.


Robert Smart (munnari!ditmela!smart@uunet.uu.net)
23 Jul 88 13:36:29 GMT


We are in learning mode about creating wide area networks. The
shortage of good solutions is shown by the NORDUNET plan. They are
going to create a wide area ethernet (via mac-level bridges) with no
hosts on it, only routers for mid-level protocols like IP and DECNET
plus funny boxes for connecting serial lines through an ethernet!

Let me say what I think is available today, and why it is deficient
for what we would like to do.

Cisco and Proteon provide boxes which will route IP and DECNET (and
other protocols of no interest to us) between ethernets. They claim
they will soon support OSI Connectionless Network routing as well.
I know a little about Proteon P4200s, and assume Ciscos are the same:
the boxes can have ethernet boards and serial line boards. However the
serial lines can only speak the secret Proteon protocol: so they have
to lead to another Proteon box.

The first problem with this is that not all points on your network can
justify a proteon box. If you have a small site with just a few Suns
or VMS Vaxes then you will want to connect them to the world by just
running slip out of one of the Suns or Asynch or Synch decnet out of
one of the VAXes. It would be nice to be able to run these serial
lines into a Proteon box serial port instead of having to run them
into some host on your ethernet. The whole point of getting Proteon
boxes is to avoid having hosts doing packet routing.

Another protocol we would like to support is X.25. It would be nice
to allow two Proteon boxes to be able to talk to each other over X.25.
Presumably this would be used as a backup mechanism for connection
when a leased line malfunctions. But it could also be used as the
normal method of moving packets between networks that don't have a lot
to say to each other. Once again there are standards for running IP
and DECNET over X.25. It would be nice to be able to run connections
from isolated IP or DECNET nodes into a Proteon box via an X.25
virtual circuit instead of having to run them through another host.

However we would also like to be able to have our router network act
as an X.25 network. I.e. host1 can talk X.25 to host2 in:

         X.25 connection private protocol X.25
    host1---------------router----------------router-------host2

Why would you want to do this? Well firstly there are plenty of
implementations now that will run OSI over X.25, so it is a natural
way to get OSI protocols now. Conversely it is not obvious when there
will be host implementations of Connectionless, or routing software
that will allow implementations of X.400 (say) over CLN to talk to
implementations connected to X.25 networks around the world. Also
there is a lot of software around that runs over raw X.25 (or via
non-standard mid-level protocols): such as the Coloured Book suite.

There are now protocols for running X.25 packets over ethernet
(pinkbook), and over DECNET (PSI-access) and now also over TCP/IP. We
would, of course, like our X.25 implementation to talk happily with
these.

Another thing that would be useful would be to have our routers
connect arbitrary synchronous lines.

  sync line----router------------router----sync line continues

This would allow for connecting IBM and other things through the
network. Obviously a sync packet entering one end should not be unduly
delayed before emerging from the other: these packets would have to
have priority. For longish packets it might well be important to
arrange for them to start emerging from the far end before they have
even finished arriving at the receiving end. In which case you better
arrange to get the rest to the far end in time!

It would also be nice if the router would act as a mac level bridge
for some or all protocols it doesn't really know about. I gather from
the nordunet proposal that this is coming.

Well, while we're at it. It would be nice to be able to use the serial
ports as a serial line concentrator. Users logging in to these ports
could be connected to their destination in many ways: telnet, rlogin,
X.29, OSI VT, or just by connecting to another port somewhere. And if
it was done nicely the user need not even be aware of how the
connection is being made. Protocol transparency: never quite comes off
in real life.

Would any of the router makers like to comment on these proposals. All
pie-in-the-sky? Is the NORDUNET wide area ethernet really the only
way?

Bob Smart <smart@ditmelb.oz.au> or <smart%ditmelb.oz.au@uunet.uu.net>



This archive was generated by hypermail 2.0b3 on Thu Mar 09 2000 - 14:42:51 GMT