morgan@JESSICA.STANFORD.EDU (08/25/88)
An additional issue with Wollongong's approach to connecting a NetBIOS LAN to IP/Ethernet is that the router software does not support a routing information exchange protocol (eg RIP), meaning lots of unpleasantness with configuring static routes to drop it into a large network. The only thing worse, I suppose, would be router software that *did* support RIP, which anyone could put up on any PC anywhere anytime with any funny network numbers they felt like . . . A better scheme, I think, would be the one used by KIP for the Kinetics FastPath: the gateway manages a set of IP host addresses from the address space of the Ethernet network to which it is attached, and responds with its Ethernet address to ARPs within that set (I guess it's just Proxy ARP, isn't it). No muss with routing issues, the PC clients run IP-over-NetBIOS as usual. Having the gateway dynamically assign IP addresses to the PC clients (ala KIP) would be even better, but not necessary. NetBIOS name discovery could handle the PCs finding the gateway's link-level address (since it's not an IP router any more). - RL "Bob" Morgan Networking Systems Stanford
dcrocker@TWG.COM (Dave Crocker) (08/26/88)
Bob Morgan is correct that our current product does not support dynamic routing information exchanges. That is slated for the next revision to the product. For the initial customer base, this does not appear to be a major problem, since the product is being used in simple networks, with few subnets. At the worst, use of a "default" next-hop is adequate. (This is a common mechanism, to simplify routing tables for leaf sub-nets, but is generally viewed as a poor, if not dangerous, long-term mechanism, which is why we are using it only for the short-term. Bob's suggestion of having a fake router which spoofs one sub-nets membership as part of another is clever but, it seems to me, frought with danger. One of the simplest concerns is that every new type of mechanism that is added to the architecture of a network alters the complexity and, therefore, predictability, of the architecture. For example, what does this scheme do to network security? Does it really make the network simpler to manage? ... This would be a convenient moment to step onto a soapbox to cite the excessive excitement people have about Level 2 (or MAC or Learning) Bridge technology, but that probably is for a different discussion group. Dave
morgan@JESSICA.STANFORD.EDU (08/26/88)
Dave Crocker writes: > Bob's suggestion of having a fake router which spoofs one sub-nets > membership as part of another is clever Hmmm, let's not assume that a device that connects between two different link-level methods has to be an internetwork router . . . > but, it seems to me, fraught > with danger. One of the simplest concerns is that every new type of > mechanism that is added to the architecture of a network alters the > complexity and, therefore, predictability, of the architecture. For > example, what does this scheme do to network security? Does it really > make the network simpler to manage? ... This would indeed be a new mechanism in the PC/NetBIOS context, but this scheme is in use by about 60 KIP gateways and over 400 Macintoshes (and some PCs on LocalTalk, too) at Stanford, and dozens if not hundreds of similar installations nation- and world-wide. Running a Kinetics box as an IP router is also possible; I believe KIP is the choice of the vast majority of K-box owners. I'll admit the idea could fail completely if the AppleTalk NBP mechanisms that are used to locate the gateway, acquire and defend addresses, etc, don't map well to NetBIOS (about which I know fairly little). Certainly it's hard to classify such a gateway as an OSI-level-N relay, but NetBIOS as a link layer is a little iffy to start with. Ease of management is in fact the major benefit of the scheme. With the KIP scheme (including dynamic IP address assignment) _absolutely_ no_configuration_ is needed to add a new user station to an existing net. No Bootp tables, no RARP tables, no IP address assignment. Of course, your organizational IP address reservation method does need to have a way to reserve a block of addresses for the gateway. And, as mentioned before, it's transparent to your IP routers as well. - RL "Bob"