smith@umn-cs.UUCP (06/06/83)
#R:vaxine:-11900:umn-cs:7400003:000:1205 umn-cs!smith Apr 25 17:07:00 1983 The Arpanet takes advantage of a key difference between itself and Usenet: the wideband lines on the Arpanet are dedicated, permanently connected lines rather than dialup lines. The routing algorithm is tied closely with software that continually measures the transmission and throughput qualities of the lines, and propagates this throughput data throughout the network once every minute. Thus, each IMP in the network maintains its own internal map of the network, including all known information about line delays. Lots of design went into mechanisms to keep these maps consistent in all IMPs in the net. Even so, packets can and do occasionally loop (for a few seconds) when the topology changes (i.e. a line goes down or an IMP goes down). I'm not sure we can apply such technology to Usenet since the speed of message propagation varies so much. Consider how much trouble we have in 'news' with message replies arriving several days before the base message. Connectivity information would be even harder to manage, or even interpret. We might be able to build something that 'usually' works. But is that good enough? Rick Smith. csnet: smith.umn-cs@Rand-Relay uucp: ..!ihnp4!umn-cs!smith