kwe@bu-cs.BU.EDU (kwe@bu-it.bu.edu (Kent W. England)) (06/09/88)
In article <1041@sun.soe.clarkson.edu> nelson@sun.soe.clarkson.edu (Russ Nelson) writes: >Reading Comer's book, _Internetworking with TCP/IP_, I get the >impression that RIP is a second son of the Internet protocol suite. I >also found out why I couldn't find any documentation on the "RIP >Protocol". Is there a move to HELLO afoot? *Should* there be a move >to HELLO? > You can read all about RIP in IDEA0004-01. As soon as Hedrick gets it published as an RFC, we can all take our respective vendors to task for any noncompliance, if there are still any implementations out there that haven't done split horizon, holddowns, etc. HELLO seems to be becoming an orphan with the emergence this summer of the "new NSFnet backbone". Are there any other implementations running, besides the "old NSFnet backbone"? What happens to the fuzzballs? Should we start a networking corner at the Boston Computer Museum? Or do they all revert to Mills' possession? :-)
Mills@UDEL.EDU (06/10/88)
Kent, So far as I know, the Hellospeak (or whatever else it may be called) routing scheme has been implemented in full only for the fuzzballs, about two dozen of which are skulking about the Internet now. Partial implementations have been done for the cisco router and gated daemon for Unix 4.3bsd. A lot of work went into the design and refinement of Hellospeak in such rugged environments as the present NSFNET Backbone, but as in many other cases like this, the sheer weight of Unix ubiquity and availability of RIP will undoubtably prevail. I would argue and expect many RIP implementors to agree that neither Hellospeak nor RIP are suitable for use on an extended WAN topoplogy and that link-state algorithms such as used in the new NSFNET Backbone are more appropriate. As for enhancements to the traditional Bellman-Ford algorithms, such as hold-down, split-horizon and so forth, there is considerable disagreement within the RIP community as to whether and how to do these things and I suspect agreement will never be reached. These things and some other tricks have in fact been done for Hellospeak, but have not been thoroughly documented and probably will not be replicated. The NSFNET Backbone fuzzbunch belongs to NSF; however, by agreement Merit & Co. is entitled to them should they choose to exercise that option. However, I doubt very much IBM wants to get into the fuzzball maintenance business, to say the least! Present plans are to pull down the serial lines at the end of July, which will leave them isolated on their local Ethers. Personally, I would like to see the fuzzies remain in place for use as remote observation platforms and as time servers for the various regional nets. It is not clear whether the regionals will agree to support this (hardware maintenance) or not. Dave