malloy@ittral.UUCP (07/18/84)
Does anybody know of, or have even heard a rumor of, some form of link from an Ethernet in one building, connecting to an Ethernet in another building. The distance is four miles, and the computers would be Vax 780 running 4.2bsd and Sun 2/170's running 4.2bsd. Thus two seperate Ethernet's but both using TCP/IP. The problem is first finding a box that will do it, and second the physical link needed. Ungermann Bass make a Bridge 1 unit which will work on lots of different links, leased lines, fiber optics almost anything you can name. However they currrently use Xerox XNS protocol and are only working on TCP/IP (for some reasons bridges seem to need to know the protocol used). Bridge Comm. also have a link, but they too use XNS and are only working on TCP/IP. Any and all information you can provide will be appreciated, even rumors of "I heard of a guy who was doing something like that". The need as is always the case, is to have it yesterday. For those people who say MOVE! You've obviously never worked for a company like ITT, it would make too much SENSE to have everyone in the same building. -- Address: William P. Malloy, ITT Telecom, B & CC Engineering Group, Raleigh NC {ihnp4!mcnc, burl, ncsu, decvax!ittvax}!ittral!malloy
wunder@wdl1.UUCP (wunder ) (07/21/84)
About linking two Ethernets two miles apart: Sun Microsystems has a fiber optic link leased from Pacific Telephone between two of their buildings. I don't know how far the link actually runs (the buildings are very close). My info is from a newspaper article, so I don't know anything more about it. At Ford, we use PDP-11/34s running DCN for IP gateways (called fuzzballs). We are going to replace them with 68K gateways (not yet finished), but those won't be a product. You can talk to Patrick Holkenbrink (patrick@ford-wdl1) about our gateways. walter underwood Ford Aerospace and Comminications, Palo Alto UUCP: fortune!wdl1!wunder ARPA: wunder@FORD-WDL1 Phone: (415) 852-4769
cak@CS-Arthur (Christopher A Kent) (07/28/84)
What you need is a *repeater*. This is a box (or in your case, a pair of boxes connected by a fiber optics link) that does bit-by-bit signal propogation between two ethernet segments. Xerox sells one for local use (that is, the two wires are close enough to each other to use normal transceiver cables to get from the repeater to each of the segments) but I don't know if they have a remote repeater (yet). DEC has announced both flavors of repeaters, but apparently is not yet delivering yet. A repeater makes it looks like you have one long cable, not two subnets connected by gateways. Broadcasts, collisions, everything propogates between the two. There is some limit on the distance for a remote repeater; you'll have to consult the Ethernet spec for details (mine isn't handy). Cheers, chris
julian@deepthot.UUCP (Julian Davies) (08/01/84)
In the original request starting this line of discussion, I seem to remember that the two ethernets were several miles apart. In this case no kind of "repeater" is appropriate, no matter if it uses fibre optics, microwave or anything else. Ethernet has a maximum end-to-end propagation delay, which is assumed predictable for determining the length of time needed to listen for contention when trying to grab the ether. The geographical distance limit is usually quoted as 1500m (say 5000 feet) on thick ethernet cable, or one third of this on 'thin' cable. Fibre-optic links etc still have the same speed of light propagation speed limit for information pulses, so the lengths of any such links must be counted towards the 1500m limit. To go over about a mile, you MUST use a full-fledged gateway or bridge which buffers each packet being transferred and rebroadcasts it on the other cable as a separate activity. It seems that quite a few organizations are using broadband cable for networks over distances of a few miles; these can be set up to operate over longer distances than ethernet, by various means, and some organizations like the flexibility of combining video with several different carriers of data. The trade-offs depend on the particular needs though. PBXs in various flavours, and local packet-switched networks (X.25 type) are also possible solutions (though the latter is fairly unusual in N. America -- but is getting quite common in the UK and maybe in Europe.) Julian Davies {deepthot|uwo}!julian
smb@ulysses.UUCP (Steven Bellovin) (08/04/84)
Julian is almost right. The Ethernet specification explicitly permits "a maximum of 1000 meters of point-to-point link anywhere in the system". Schematically, this is a single repeater with a *very* long mid-section. The diagram below is an attempt to reproduce a picture in the 1.0 Ethernet specification. The asterisks represent hosts, the periods are drop cables, and the pairs of parentheses connected by lines are repeaters. Each of the coaxial cable segments can be 500 meters, thus permitting an end-to-end distance of considerably greater than one mile. To be precise: the absolute upper bound can be calculated by adding together the maximum cable lengths, including the controller-transceiver drop cable of 50 meters. If everything shown below was connected at the appropriate end of the cable, the distance between hosts 1 and 2 could be 50 (drop cable from host 1) + 500 (cable A) + 50 (drop cable to repeater) + 50 (drop cable from repeater) + 500 (coaxial segment C) + 50 (drop cable to point-to-point link) + 1000 + 50 (drop cable from point-to-point link) + 500 (coaxial segment D) + 50 (drop cable to host 2), or 2800 meters. A B 1 *..| |..* | | |..* *..| | | *..| |..* |...(-) * * (-)...| . . . . . . . . ----------------------------------------- C . . (---------------| | | | | 1000 meters D | | | | |..* | | |-----------------).....| | |..* | 2 *..| Now, I'm not claiming that you can use this to connect two buildings a mile apart -- but you can do far better than 500 meters. (Those who wish to calculate the true worst-case round-trip time should read the spec. It's rather more complex than one would think, since there are different figures for signal speed in assorted media (.77c for coax, .65c on the drop cables, etc.), startup times for encoders and repeaters, signal strength considera- tions, etc. The final answer, according the 1.0 standard (my copy of the 2.0 standard is in my office) is 44.99 usec. --Steve Bellovin
kissell@flairvax.UUCP (Kevin Kissell) (08/05/84)
With regard to distance limitations on Ethernet, they are defined
in order to provide guarantees of collision detection. (There are
electrical considerations as well, but they are less restrictive.)
Violating distance spec will result in an increase in the probability
of a collision going undetected, and thus of a damaged packet being
recieved. So long as higher level protocols provide for error recovery,
a network can still run. The increase in retransmissions is very small
so long as cable utilization remains relatively low. Losing collision
detection *can* hurt you badly if you're using up 90%+ of your bandwidth.
Now, I'm not saying that the Ethernet distance specs should be treated
with the same disdain that many people show for the 50' RS-232 distance
restriction, but I recall hearing about an experiment at ITT Belgium
where they ran some Ungermann-Bass gear down several miles of Ethernet
cable and still got what they considered to be very acceptable system
performance. Of course, you break the rules at your own risk...
Kevin D. Kissell
Fairchild Research Center
Advanced Processor Development
uucp: {ihnp4 decvax}!decwrl!\
>flairvax!kissell
{ucbvax sdcrdcf}!hplabs!/
"Any closing epigram, regardless of truth or wit, grows galling
after a number of repetitions"