jim (04/14/83)
[ I am moving this discussion, since it doesn't really belong in net.news.group. ] On the problem of replying to net news: I gathered together from my uucp log file a list of addresses for all mail passing through uw-beaver for the last 30 hours, not including mail originating here. There were 59 messages, of which 17 were going more than one hop away. Of these 17, 8 had less than optimal paths, two had very bad paths (7 and 6 hops where 2 and 1 would have done), and one had a better path than our routing tables. The other six had paths different from what our routing table would use, but it was not clear whether they were better or worse. I have considered fixing our uucp mailer to redirect mail to the last machine in the path that our routing tables know about. For example, mail addressed to cornell!vax135!ariel!houti!hogpc!houxt!4341shn would be redirected to allegra!houxt!4341shn Has anyone actually implemented anything like this? Can you think of any problems with this scheme? It makes "bounce tests" impractical, and puts a large burden on the system administrator to keep the routing tables up to date, but in some cases it would greatly improve message paths.
smb (04/15/83)
There are serious problems with attempting optimal routing of netmail passing through. For one thing, loops are possible if more than one node does this. (I route all my long-distance calls to you; you route them right back to me.... It isn't clear to me whether or not such loops can occur if everyone uses the same routing tables.) For another, loop tests won't work properly. Finally, deliberate attempts to route around a node known to be bad will fail if your routing tables think it's working. --Steve