csg@pyramid.pyramid.com (Carl S. Gutekunst) (06/02/88)
I'd like to make a plea with all sites to stop listing Internet connections on their UUCP Maps. The problem is that it is getting increasingly difficult for me to generate meaningful pathalias output. I can twiddle the numbers to make the paths come out right, but this is becoming tedious. In this day of domains and smail, there is simply no good reason for any site to list their Internet connections on the UUCP Maps; domains handle the routing. Isn't this the whole reason domains were created, to simplify the maps? The worst "offenders" for us are Rutgers and Ames; I have discussed this with their respective administrations and they felt they wanted the map entries kept as is. But why? I mean, there is this nice bunch of lines in D.top: rutgers .uucp rutgers .arpa, .com, .gov, .mil, .edu, .org, .net, .us rutgers .de, .no, .nz rutgers .bitnet That tells me everything I need to know about rutgers connections. Given this, why does any site need to show explicit names in their list of links? Comments? Mel? <csg>
david@ms.uky.edu (David Herron -- One of the vertebrae) (06/02/88)
Carl, I have read through this 4 or 5 times and still cannot figure out exactly what you are complaining about. What do you mean by "Internet connections"? In article <25550@pyramid.pyramid.com> csg@pyramid.pyramid.com (Carl S. Gutekunst) writes: >... The problem is that it is getting increasingly difficult for >me to generate meaningful pathalias output. I can twiddle the numbers to make >the paths come out right, but this is becoming tedious. eh? why do you need to twiddle anything? Maybe if you give an example of what sorts of things you're having troubles with? >In this day of domains >and smail, there is simply no good reason for any site to list their Internet >connections on the UUCP Maps; domains handle the routing. Isn't this the whole >reason domains were created, to simplify the maps? Well. That's not the real reason that domains were created, but it *was* the reason we push domains for the uucp world so I'll let it stand as is. >The worst "offenders" for us are Rutgers and Ames; I have discussed this with >their respective administrations and they felt they wanted the map entries >kept as is. But why? Well, I was curious and took a quick look in the map to see what was going on. Do you perhaps mean the section of stuff listed as "TCP/IP" connections? What's your problems with that? If all those connections are anything like the one they have with us it's done with real UUCP over TCP connections across the internet. Regardless of how it's done I'm sure it'd be invisible to you so what's the problem? >I mean, there is this nice bunch of lines in D.top: > > rutgers .uucp > rutgers .arpa, .com, .gov, .mil, .edu, .org, .net, .us > rutgers .de, .no, .nz > rutgers .bitnet > >That tells me everything I need to know about rutgers connections. Given this, >why does any site need to show explicit names in their list of links? how? If you take away all of the explicit names then how are we supposed to generate routes within the net? If I understand what you're suggesting correctly the only thing in the site entry would be the comment section. Now, I do have a problem with rutgers' entries in general. In u.usa.nj.1 they have a number of entries for tiny machines (3b2's). They don't need to do this -- they can easily cause the mail systems on those machines to generate domain type addresses EVEN if the machines are uucp only. smail is wonderful software and does good things. You don't even have to have much of a database on the local machine -- my home machine has ONE entry, the one giving the route to "ukma" the local smart-host. >Comments? Mel? > ><csg> -- <---- David Herron -- The E-Mail guy <david@ms.uky.edu> <---- s.k.a.: David le casse\*' {rutgers,uunet}!ukma!david, david@UKMA.BITNET <---- <---- Goodbye RAH.
pleasant@porthos.rutgers.edu (Mel Pleasant) (06/02/88)
Um, not *all* of the rutgers systems are under my control. I can't
_make_ them run a smart mailer.
--
Mel Pleasant
{backbone}!rutgers!pleasant pleasant@rutgers.edu mpleasant@zodiac.bitnetnowicki%rose@Sun.COM (Bill Nowicki) (06/03/88)
In article <300@ncar.ucar.edu>, woods@ncar.ucar.edu (Greg Woods) writes: > > ... Not everybody in > the world is running MX-compatible mailers yet. To get one here, I had > to search high and low for a hacked version of sendmail 5.58 that would > run on a Sun-4. It took weeks to find it. Not everyone has that much > time to spend on something like that. Note that an MX version of sendmail is included in the current software release (SunOS 4.0). This has been shipping to thousands of customers for about a month. -- Bill Nowicki Sun Microsystems
mcb@lll-tis.UUCP (06/03/88)
I'm beginning to get a glimmer of what Carl means in the "Second try",
but I still don't think I agree with the underlying assumptions.
Assuming that everyone is being honest (or at least rational) about
the pathalias cost of their links, both UUCP and Internet, then we
must assume that in the absence of pathological behavior the least
cost (i.e., "most efficient") routes are chosen.
One problem is that the pathalias constants (at the lower numeric levels;
you don't see too many "WEEKLY" entries any more) are stated as "grade of
connection" ("DIRECT", "DEDICATED") but people attempt to derive cost
and speed of delivery from them.
In many cases multiple hop SMTP service is *much* quicker than
single-hop UUCP service, and for people whose network costs are not
(yet) usage-sensitive (hint, hint), it is cheaper as well.
We only list internet connections in the map entry when we have a
TCP-UUCP connection with the site as well, so that mail to
<us>!foo!bar
will succeed if we have a TCP-UUCP connection to foo.gov, but
bar@foo
will fail since foo is not in our local domain;
<us>!foo.gov!bar
is handled by the sendmail config and will succeed in any case.
Michael C. Berch
mcb@tis.llnl.gov / {ames,ihnp4,lll-crg,lll-lcc,mordor}!lll-tis!mcbcsg@pyramid.pyramid.com (Carl S. Gutekunst) (06/03/88)
Well, every followup and mail reply I've gotten has misunderstood what I was proposing, so it must have been pretty awful. (I dunno, made sense to me. :-)) Let me try again. And please, don't construe this is a flame at Rutgers or Ames; they're two of our best links, and I respect Mel Pleasant's judgement a *lot*. I just want to raise an issue that I think is more important than the latest about JJ, and see what other SysAdmins think. First, I'm only talking about eliminating sites, not domains. Adding domains to the UUCP maps has worked miracles in connectivity. Second, I'm only talking about eliminating sites on fully connected networks, that are directly accessible from hundreds of sites on the Arpanet or Milnet. Sites that are in the hosts table, so MX records aren't an issue. (I'll bet Greg lunch that there aren't more than a dozen sites in the UUCP map that have Arpanet or Milnet access and aren't in the host table. See you at USENIX? :-)) What I'm trying to avoid are two types of pathalias funnies: paths that take long distance UUCP hops when a local call would do, and paths that prefer a multiple-hop Internet link when an adjacent UUCP link would do. Take princeton (aka princeton.princeton.edu) as an example. I would like to be able to route to them via decwrl, sun, ames, or one of the other local Arpanet or Milnet sites of my chosing. But because princeton is in rutgers' UUCP Map, and we give rutgers a low cost because it is an important link, pathalias wants to route to princeton via rutgers. As an example of the reverse, we have a direct UUCP link to ut-sally. Ames lists ut-sally(DEDICATED) on their UUCP Map, so pathalias choses the call to ames and a hop on the Milnet (through the Mil/Arpa gateway?) instead of the direct UUCP link. The only way to avoid this is to inflate the cost for ames, or use an unnaturally low cost for ut-sally. (I submit that this problem is much trickier, and certainly much more arguable. I mean, the hop via ames honestly is the cheapest.) The alternatives are to either list all Internet sites in the maps for every site that has Internet access, or list none of them and let the (preferably top-level) domains handle the routing. The former is impractical; there are almost 6500 nodes in the latest host table. That's why I'm proposing the latter. An additional possibility would be for Arpa or Milnet sites that have UUCP map entries to list a reasonable cross-section of other Internet/UUCP gateway sites in *their* maps; for example, the princeton map would contain: decwrl .princeton.edu rutgers .princeton.edu ucbvax .princeton.edu or something similar. In fact, I am doing this locally for domains that I am sure are physically on Arpa or Milnet; but that does nothing for our neigh- bor's paths, and still leaves the Internet-overrides-UUCP routing problem. Am I making sense now? Can we try this again? <csg>
honey@umix.cc.umich.edu (Peter Honeyman) (06/05/88)
i run pathalias output through this script: egrep -v '(\.(com|edu|mil|gov|net|org|arpa|[a-z][a-z]) .*!.*!)|(.\.(com|edu|mil|gov|net|org|arpa|[a-z][a-z]) )' peter