andrew@garfield.UUCP (Andrew Draskoy) (12/09/83)
A little while ago I was running mkpath on Rob Kolstad's uucp
routing data and discovered that there were two systems out there
who seemingly talked to garfield, that we at garfield didn't know
anything about. Well, the net being what it is, this wasn't really
surprising. I sent myself mail through them to make sure that they
really did talk to us, and ran into the following weirdness:
Akgua, which is apparently AT&T/Bell Labs/Whoever-they-are-now's
latest I-talk-to-everybody site does indeed call us. That's ok, we
certainly won't object! The second site, linus, is where the
bizarreness comes in: If you mail to linus!garfield!user, linus
says "Ah yes, garfield... I just send this through allegra" and the
path gets changed to linus!allegra!garfield!user. Well, it's great
if linus knows what to do with mail to garfield, but it doesn't
really have a uucp connection to us, so I'm not so sure that it
should advertise that fact that it does. I think perhaps optimization
of routes for mail that comes through linus towards us would be a better
idea. Suppose for instance, that someone who can call system foo,
which talks to system bar which in turn talks to garfield, for $0.10
and linus for $0.20 wants to send me mail. Presumably it will send
through the cheapest system (foo) if the number of hops to garfield
is the same. (This is easy to set up with mkpath, just give it your
directly connected systems in reverse order of cost.) But the number
of hops through linus is (it thinks) smaller, so the mail goes through
linus. In fact the number of hops was the same, and someone just
lost $0.10. All of this is besides the important question: Do we
really want to add yet another level of comlexity to uucp routing?
Or is it new? I haven't run across it before.
I'd like to hear other people's views on this, especially anyone who
knows what's going on (at linus, perhaps?)
One last note: Has anyone else been added to akgua list of connections?
We don't have an L.sys entry for them, and were wondering about getting
one. The problem is that there is no answer from akgua!{postmaster,usa},
the later being the contact for akgua listed in net.news.map.
--
Andrew Draskoy
{akgua, allegra, ihnp4, linus(sort of), utcsrgv}!garfield!andrewsmk@linus.UUCP (Steven M. Kramer) (12/10/83)
Ah, yes. It's not bizarre! There are quite a number of sites
logically around us that don't have the time or desire to implement
a database of uucp sites. Steve Hemminger, while he worked here this
year (Hey, Steve, how's it going?) implemented smb's pathalias program
and had it so that we are virtually connected to any site that is
in our database. Other sites aound us can mail to linus!<site>!<user>
and get their mail there. I feel this is much better than stopping
the mail.
--
--steve kramer
{allegra,genrad,ihnp4,utzoo,philabs,uw-beaver}!linus!smk (UUCP)
linus!smk@mitre-bedford (MIL)ka@hou3c.UUCP (Kenneth Almquist) (12/13/83)
I applaud linus's efforts in respect to mail delivery. One word of caution, though: path expansion can lead to infinite loops. Infinite loops will not occur if data bases are kept consistant, but if many sites begin providing path expansion, I think that inconsistant data bases are inevitable. Therefore, sites which provide path expansion should check for infinite loops. Kenneth Almquist
smk@linus.UUCP (Steven M. Kramer) (12/15/83)
Thank you, Kenneth A. I have seen what could amount to infinite loops
long ago, when a mail path simply went bwyond the small (100 chars) buffer
size. It wasn't due to anything like our routing -- sinply was a long
path. What I noticed was that uuxqt would always bomb there. I guess
this is a sufficient test, although certainly a better one could be
done, like a double detection of something passing thru a particular
site.
--
--steve kramer
{allegra,genrad,ihnp4,utzoo,philabs,uw-beaver}!linus!smk (UUCP)
linus!smk@mitre-bedford (MIL)ggr@pyuxbb.UUCP (12/17/83)
If your site uses sendmail, it detects what it thinks are loops by counting the number of "Received: " lines in the header. The message is returned if the number exceeds 30 (I think that's the number). If the error message causes a loop, eventually somebody's "Postmaster" or "MAILER-DEMON" gets the message. === Guy Riddle == AT&T Bell Laboratories, Piscataway ===
laura@utcsstat.UUCP (Laura Creighton) (12/18/83)
You know, I wouldn't want to carry 30 "Received:" lines around with me whenever I am foolish enough to try to mail something deep within the bowels of Holmdel... I always thought that the "Received:" lines were debugging options, not a feature that sendmail is relying upon. If huge headers which fill a screen are a "feature", maybe we should reclassify them as a "bug" and figure out something else...like walking back the ">From" lines and seeing if it has been here before. Laura Creighton utzoo!utcsstat!laura
jsq@ut-sally.UUCP (John Quarterman) (12/19/83)
There is no real reason not to carry 30 Received: lines around with you.
They have little effect on the speed of transmission of the message
and any mail reading program that does not allow for the suppression
of display of unwanted header lines is defective.
--
John Quarterman, CS Dept., University of Texas, Austin, Texas
{ihnp4,seismo,ctvax}!ut-sally!jsq, jsq@ut-sally.{ARPA,UUCP}mark@cbosgd.UUCP (Mark Horton) (12/20/83)
"Received:" lines are not for debugging, they are intended to be used in all mail (not just sendmail, anything conforming to RFC822) as a trace of where the message went. That way, when something goes wrong (not IF, WHEN, as it eventually will when it crosses into some weird domain or piece of software) it is possible to look at the headers to tell what happened. It also is a good indication of the route your mail took to get to you and how long each hop took. If you don't like reading them, fix your user interface, don't violate the standards. In particular, 4.2BSD's Mail program has an option to suppress any headers you don't want to see. My .mailrc contains the line ignore received status via message-id date sent-by which cuts out the boring stuff. If I really want to see all this stuff I use the P command. By the way, netnews does something similar in not showing you all the headers on the article, since they are also longer than you are likely to want to read. Mark