Path: utzoo!mnetor!uunet!ksr!ubglue!dudek From: dudek@ubglue.ksr.com (Glen Dudek) Newsgroups: comp.mail.uucp Subject: Re: Rerouting of explicit paths Message-ID: <261@ksr.UUCP> Date: 12 Mar 88 21:52:30 GMT References: <327@vsi1.UUCP> <202@ditka.UUCP> Sender: nobody@ksr.UUCP Reply-To: dudek@ksr.UUCP (Glen Dudek) Organization: Kendall Square Research, Cambridge MA Lines: 90 In article <202@ditka.UUCP> kls@ditka.UUCP (Karl Swartz) writes: >One of the guys at formtek (my office) had a friend who's sole >connection to the world was via harvard. Apparently, harvard >would either auto-route or send everything to rutgers to auto- >route. I was postmaster@harvard.harvard.edu from 6/85-6/87. My policy has always been to *only* look at the first host in a UUCP path, and only autoroute if we didn't talk (or didn't want to pay the long distance to talk) directly to that host. The only mail that is directed to rutgers is mail that was explicitly addressed there, or mail for which the next hop was through a host which pathalias thought was best reachable through rutgers (e.g., most ATT sites). I know for a fact that this software has not been changed to any significant degree since I left (KSR's major mail link is through harvard, surprise :-). In article <479@minya.UUCP> jc@minya.UUCP (John Chambers) writes: >Around here, we have a lot of very similar problems with mail that >stops off at harvard. I get lots of mail that came to me via paths >like ...!harvard!mit-eddie!harvard!foo!harvard!bar!adelie!minya!jc. >The nice thing here is to note that mit-eddie!minya is a uucp link. >The map entries show both that link and adelie!minya to be DAILY, >so it's not clear why eddie bounces it back to harvard. The last time I looked, putting together an up-to-date pathalias database was a large investment in time and effort. Getting one that was usable was even more so! By usable I mean one that generated workable paths for most sites. There was no good way to verify that a pathalias database was a 'good' one other than by gross inspection of some of the more commonly utilized paths. One of the more successful methods used by Nike Horton was to generate a count of paths/first hops, (e.g. 4000 hosts have 'rutgers' as the first hop in the path), and to examine this table to see if it matched expectations. Expectations, of course, were learned only through experience. Experience, of course, was learned only through mistakes. This procedure is probably better now, but the point is that it cannot be verified. In any case, the problem John Chambers writes about is probably due to out-of-date or conflicting pathalias data at least between harvard and mit-eddie. Harvard probably thinks mit-eddie!minya is the best path, since it uses SMTP to get to mit-eddie and only UUCP to get to adelie. Mit-eddie may think adelie!minya is better than its own direct UUCP to minya, and mail to adelie should (correctly) go through harvard. Perhaps the postmaster@mit-eddie eliminated the link or doesn't want to overutilize it. How is harvard to know until/unless the maps are updated? What will happen to the mail during the fuzzy in-between time while the maps are being updated if auto-routing is used? I don't even want to think about it. >...I've brought >it up with those machines' postmasters, but they are very busy. This is very true. The only reason I was able to maintain mail in a reasonable state was because I dedicated a large amount of time over 6 months to understanding the details of mail and sendmail. I know the current staff at Harvard is undermanned, and I am sure they are reluctant to dedicate the time necessary to have the confidence to make the mistakes necessary to learn the right changes to make. If you are using Harvard for auto-routing, you are relying on them to maintain their data at least as up-to-date as you would. If they do not, then you can maintain it yourself, and count on Harvard not to rewrite the path if you need to send mail through them. >...Eventually we'll probably have some self-installing and >self-correcting packages. For the present, when you submit your >mail to the network, it is ending up in the jaws of lots of semi- >tamed mailers whose behavior is as puzzling to the owners of the >machines as it is to you and me and other mere humans. This is also very true. I can understand why rutgers wants to autoroute, but I can't approve of until mail systems are self-installing, self-correcting, and self-maintaining. The UUCP maps are still bound by GIGO, and GI is still the rule. The regional map maintainers do their damndest, to their great credit, but they cannot completely verify the correctness of the UUCP data they get. >If we have patience, but keep harassing the culprits who foist these >"intelligent" mailers on us, maybe eventually we'll have something >that works. I think 'notifying them of problems' is more appropriate than 'harassing', and try and make sure you take the trouble to find the right culprits. Anyone in a UUCP path can rewrite it - it doesn't have to be harvard or rutgets. We really do need mail headers and log info which records the incoming and outgoing From: and To: addresses. How about something similar to the IP record-route option? Glen Dudek ksr!dudek@harvard.harvard.edu