OWEN@AUDUCVAX.BITNET (Larry Owen) (04/12/88)
Here's a real rookie question:
I know (or at least am laboring under the assumption) that all hosts
(connections) on a given physical network must (should?) use the same
network number in their internet address. What about MAC-level bridges
(Lan Bridge 100's, Vitalink Translans, and the like)? Are all ethernet
segments connected by these devices on a single net (for purposes of
internet addressing), or can each segment use a different network
(or subnet) number?
Thanks in advance,
Larry Owen
Academic Computing Services
Auburn University
OWEN@AUDUCVAX.BITNETmap@GAAK.LCS.MIT.EDU (Michael A. Patton) (04/12/88)
Date: Mon, 11 Apr 88 18:30 CST From: <OWEN%AUDUCVAX.BITNET@CUNYVM.CUNY.EDU> (Larry Owen) X-Original-To: tcp-ip@sri-nic.arpa, OWEN ... I know (or at least am laboring under the assumption) that all hosts (connections) on a given physical network must (should?) use the same network number in their internet address. ... This is not true. We have an Ethernet here with hosts on both 18.0.0.0 and 128.30.0.0 without (many) problems. However, the two groups of hosts do not talk to one another. This could be arranged in several ways, but we don't have any need. There is no conceptual reason why one physical network cannot be several logical (sub-)nets. Mike Patton Network Hacker MIT LCS
Doug_Nelson@UM.CC.UMICH.EDU (04/13/88)
Larry, to answer your question, a set of ethernet segments connected by MAC level bridges is indeed one logical network for the purposes of internet addressing. We are running a large class B+ network with 20 or so MAC bridges, spanning our entire campus. Running the network this way certainly simplifies our local routing problem. It has its own headaches, but there are other benefits, such as being able to run Decnet, etc., over the same wire. Doug Nelson den@serv1.cl.msu.edu Computer Laboratory 08071den@msu.bitnet Michigan State University
jqj@HOGG.CC.UOREGON.EDU (04/13/88)
>We have an Ethernet here with hosts on both >18.0.0.0 and 128.30.0.0 without (many) problems. One must be very careful when running multiple IP networks on the same cable. For example, it makes it much more likely that a misbehaving gateway will cause a broadcast storm aka "meltdown." Typical example: suppose that 128.30.0.0 is 8-8 subnetted, and that subnet 128.30.3.0 is on the shared cable. Suppose also that both IP networks have gateways to the arpanet or something. Suppose some host on 128.30.3.0 sends an IP broadcast (as an Ethernet broadcast) to its subnet broadcast address, say to 128.30.3.255. A gateway between 18.0.0.0 and the arpanet, say 18.0.0.1 (but NOT also configured to be on 128.30.3.0), receives the packet, notes "this is a packet for host 3.255 on net 128.30.0.0" (please observe that since this gateway does not itself have any interfaces on 128.30.0.0, it doesn't know what the subnet mask of that network is and hence can't recognize this as a subnet broadcast), sends it to a gateway on 128.30.3.0, who explodes it onto the cable. The first gateway gets another copy, forwards that packet, etc. As Chuck Hedrick noted in a recent message to tcp-ip, it is essential that gateways pay attention to whether the packet arrived as a media broadcast; unfortunately, many gateways (e.g. 4.3BSD) do not. A fortiori, unless you're willing to also have gateways drop ALL packets sent as Ethernet broadcasts (which I am not), or are willing to insist that any gateway on a physical network know about all the nets on that cable (which I don't see how you can do), you must also forbid topologies such as the one above.