mcm@ncsu.UUCP (08/04/83)
I also have mail routing programs, and I agree with Steve Bellovin. I originally wrote mkpath because I had an 11/60 and couldn't use Steve's pathalias program due to memory restrictions. His programs are the best to use if you can use them. The problem is with incorrect data. The data Mark Horton has seems to be everyone's L.sys info, rather than what sites they poll. That is why "-2" is an option on mkpath, instead of the default. To generate optimum paths, Mail routing programs need to know who you poll and how often. Often it is better to go through two sites that connect regularly than to go through one site that connects "once in a blue moon". If the connection list Mark Horton has contains sites that only connect once in a while, OF COURSE my mailer router will generate non-optimal paths. I've been thinking of changing mkpath so that it only takes the information from the "NEWS:" field. The "MAIL:" field for the most part is a joke. Mike Mitchell duke!mcnc!ncsu!mcm
mark@cbosgd.UUCP (08/05/83)
I'm seeing all sorts of discussion in here saying all sorts of things, some of which resemble reality. Let's set the record straight. Steve's pathalias programs .... are the best to use if you can use them. The problem is with incorrect data. The data Mark Horton has seems to be everyone's L.sys info, rather than what sites they poll. That is why "-2" is an option on mkpath, instead of the default. I don't know why you think you know what's in my database. In the first place, I think you're talking about the Usenet directory (which Karen does almost all the work on - it's HER database), which has little or nothing to do with mail. All that database does is show NEWS connections. It also has a few pieces of mail information, but the mail info there is not really useful, and the degree of work involved to make it useful is more than we have time to do out of the goodness of our hearts. I in fact DO have a mail routing database, but my current copy is not particularly organized or up to date. I'm one of many that need someone to keep the database up to date, as Karen has proposed to do (for a price). To generate optimum paths, Mail routing programs need to know who you poll and how often. Often it is better to go through two sites that connect regularly than to go through one site that connects "once in a blue moon". If the connection list Mark Horton has contains sites that only connect once in a while, OF COURSE my mailer router will generate non-optimal paths. My data is based on Steve Bellovin's pathalias program's format, and thus takes costs into account. Karen is proposing to use this format as well. Such an "official connections list" is in fact already being distributed monthly by Karen & Mark Horton, viz. in the form of a Usenet map. It's quite easy to retrieve a pahtnames database from that. Again, what we're distributing is NOT an official MAIL connections list, it's an official NEWS connections list. The mail connections are far far more involved - there are perhaps 20 or 50 times more active mail links than news links. One other thing I'd like to make clear. Karen is not an employee of Bell Labs. The service she has proposed would not be run by Bell Labs, nor would it be run using the cbosgd machine (which is owned by Bell Labs). It would be a small business run independently from AT&T, BTL, or myself. BTL has generously allowed her to use the cbosgd machine for the Usenet directory, from which the entire community have benefitted. But obviously a business has to be run from its own machine. I'd also like to emphasize that, while I think it's great that she's considering starting a business, she's doing it all on her own. I am not part of this business - I work for BTL. One other point - I have seen a lot of people say that what is needed is for someone to keep a master copy of the UUCP directory on line. I agree. But what everybody seems to expect is that some sucker is going to offer to do this for free. We've gone about a year now and nobody is keeping an up to date UUCP map. Nobody has offered. If someone is willing to do that much work for free, I think it's great. I'll wholeheartedly lend them my encouragement. But I think it's time to realize that if nobody is going to do it for free, it isn't going to get done unless somebody gets paid to do it. Karen has done a lot of work for Usenet for free, and I for one really appreciate it. I think the rest of the net does too. But you have to realize that just because she's doing a 5 hour per week job for free as a volunteer does not obligate her (or me) to do a 40 hour per week job for free. Mark Horton
thomas@utah-gr.UUCP (Spencer W. Thomas) (08/06/83)
Well, now, wouldn't it be nice if the polling frequency info were in the L.sys file, instead of, for example, crontab? Or even worse, there are those sites which manually dial the phone for their uucp connections! The frequency of polling info is in somebody's head there. =Spencer
ptw@vaxine.UUCP (P. T. Withington) (08/08/83)
Has anyone looked into adapting either steve's or mike's routers to work "on-the-fly" by "improving" a path found in mail in transit. It seems that such a mechanism would lessen the impact of a poor database. 't` --Tucker (ptw@vaxine.UUCP) ~
smk@linus.UUCP (Steven M. Kramer) (08/09/83)
It's hard to improve a site. To go from a->d, it may be quicker to go from a->b->c->d if the polling frequency is much quicker. However, I would suggest an optimizer that simply redid a a->b->a->c to a->c (i.e., eliminate loops). Any other optimization, it was pointed out some time ago, could cause looping if the DB on another site redid the optimization another way. -- --steve kramer {allegra,genrad,ihnp4,utzoo,philabs,uw-beaver}!linus!smk (UUCP) linus!smk@mitre-bedford (ARPA)
lda@clyde.UUCP (08/09/83)
Changing the mail routing database 'on-the-fly' would be great if you could be sure that people never made mistakes in the addresses (paths) on their letters. This is not a safe assumption. --- Larry D. Auton Bell Labs Whippany NJ (201) 386-2907
ka@spanky.UUCP (08/10/83)
Eliminating loops in mail addresses doesn't make much sense since I suspect that the only time you see loops is when somebody put them in specificly for testing purposes. Spanky's data base lists a total of 985 paths. Of these, 556 or 56% are direct or include one intermediate system, so that no optimization will help. The longest paths include 5 intermediate systems; there are only 3 paths of this length. Kenneth Almquist