lyndon@aurora.AthabascaU.CA (Lyndon Nerenberg) (04/14/89)
Who (if anyone) is acting as the root server for .CA ? It appears that
noone advertises an MX for .CA; I thought that all top level domains
connected to the Internet required this.
I was also a bit surprised to discover there is no "meta gateway" for
.CA listed in the maps. These do exist for .edu, .com, etc., why not
one for .CA ?
(I ask this because domains were supposed to solve a lot of problems
w.r.t mail routing. After convincing a number of local users to use
domain addresses they are beginning to think the whole concept quite
rude when domain addr's return "no such host" but old fashioned bang
paths still work.)
--
Lyndon Nerenberg Computing Services Athabasca University
{alberta,attvcr,ncc}!atha!lyndon || lyndon@nexus.carayan@AI.TORONTO.EDU (Rayan Zachariassen) (04/14/89)
In article <520@aurora.AthabascaU.CA> lyndon@aurora.AthabascaU.CA (Lyndon Nerenberg) writes:
# Who (if anyone) is acting as the root server for .CA ?
relay.ubc.ca is the primary,
clouso.crim.ca, nnsc.nsf.net, and bay.csri.utoronto.ca are presently
the secondary authoritative servers.
# It appears that noone advertises an MX for .CA; I thought that all top
# level domains connected to the Internet required this.
Right. Nope. See below.
# I was also a bit surprised to discover there is no "meta gateway" for
# .CA listed in the maps. These do exist for .edu, .com, etc., why not
# one for .CA ?
Actually, there were a few listed, namely the U.S. UUCP->Internet gateways.
I (re-)discovered this a few days ago and asked for them to be removed.
They serve no purpose because *all* .CA domains are listed with explicit
routes in the pathalias data, so if it isn't in the map it doesn't exist.
Hence there is no point sending it to a gateway like (e.g.) uunet, because
they won't know any better than you do.
What you are expecting is commonly called "topological routing". You
have to remember that the namespace is completely orthogonal to the
routing problem; think of domain names as uucp node names and you will
have a good idea of how names relate to routing (they don't). Usually
topological routing is used at the point where the namespace hierarchy
closely resembles/overlaps the network topology, e.g. at organizational
gateways. If one did this at the country level without a corresponding
network topology, you would put unwarranted strain on one or a few gateways
instead of distributing the traffic. Ideally there should be NO change
in traffic pattern just because a site converts to using domain names.
# (I ask this because domains were supposed to solve a lot of problems
# w.r.t mail routing. After convincing a number of local users to use
# domain addresses they are beginning to think the whole concept quite
# rude when domain addr's return "no such host" but old fashioned bang
# paths still work.)
I can't think of any reason for it to not work unless if the subdomain in
question doesn't properly support their own name. If there is a problem
with the maps or the forwarding mechanisms on the various nets, let me
know the details and we can investigate.
The reason that domains were supposed to solve a lot of problems were
primarily:
- they specify absolute addresses, no more source routes or bang paths
unless you or your mailer have good reason.
- the organization-internal routing can be hidden from the outside world,
which means one only needs routing information for the *sub*domain as
a whole (in terms of its gateways) rather than all internal hosts.
Wildcard routes to .CA are indeed used to reach NetNorth and CDNnet's gateways
to the outside world, but that is because they both operate a single gateway
to the world so all traffic has to go through it anyway. The Canadian
UUCP network probably has 20 links to the U.S., so topological routing
is not reasonable, nor is it necessary since any gateways would have no
better routing knowledge than any other site using the UUCP maps.
rayanlyndon@cs.AthabascaU.CA (Lyndon Nerenberg) (04/15/89)
In article <89Apr13.220817edt.38140@neat.ai.toronto.edu>, rayan@AI (Rayan Zachariassen) writes: >In article <520@aurora.AthabascaU.CA> lyndon@aurora.AthabascaU.CA (Lyndon Nerenberg) writes: ># I was also a bit surprised to discover there is no "meta gateway" for ># .CA listed in the maps. These do exist for .edu, .com, etc., why not ># one for .CA ? > >Actually, there were a few listed, namely the U.S. UUCP->Internet gateways. >I (re-)discovered this a few days ago and asked for them to be removed. >They serve no purpose because *all* .CA domains are listed with explicit >routes in the pathalias data, so if it isn't in the map it doesn't exist. >Hence there is no point sending it to a gateway like (e.g.) uunet, because >they won't know any better than you do. The problem is, there will always be times where the .CA root servers (not really the right name for them, but ...) hold information that doesn't match the published routing tables (in comp.mail.maps). In order for the above scheme to work reliably, you have to a) Ensure the new routing tables are distributed reliably in minimum time, and b) Ensure the site admin's update their mailers with the new tables in minimum time. If someone misses a map posting due to a transport problem, and if that posting contains gateway information for a new subdomain, that site (at the least) will not be able to find a route to the new subdomain until the next map posting comes along (at least a month). I also doubt that the majority of sites run pathalias on a daily basis to stay up to date on current routing. A lot of people I've talked to/worked with collect updates for two to four weeks before running pathalias. At the extreme end of the spectrum, when I started work at Athabasca University I found that the last pathalias run had taken place over a year before my arrival :-) I condone none of the practices mentioned in this paragraph, however they do tend to reflect reality. What is so evil about listing the root server in the maps as an absolute worst case gateway to .CA ? If the subdomain information is as up to date as you claim, the path will hardly be used anyway :-)
molnar@gpu.utcs.utoronto.ca (Tom Molnar) (04/15/89)
In article <524@aurora.AthabascaU.CA> lyndon@nexus.ca (Lyndon Nerenberg) writes:
# What is so evil about listing the root server in the maps as an
# absolute worst case gateway to .CA ? If the subdomain information
# is as up to date as you claim, the path will hardly be used anyway :-)
Why do you think that an MX record will exist for all sites
registered under .CA? There is no requirement for anyone registering
under .CA to have an Internet forwarder.
--
Tom Molnar
Unix Systems Group, University of Toronto Computing Services.rayan@ai.toronto.edu (Rayan Zachariassen) (04/15/89)
In article <524@aurora.AthabascaU.CA> lyndon@nexus.ca (Lyndon Nerenberg) writes:
# that posting contains gateway information for a new subdomain, that
# site (at the least) will not be able to find a route to the new
# subdomain until the next map posting comes along (at least a month).
...
# I also doubt that the majority of sites run pathalias on a daily
# basis to stay up to date on current routing. A lot of people I've
# talked to/worked with collect updates for two to four weeks before
# running pathalias.
Subdomains should not expect to be able to use their domain name the instant
it gets approved. It takes time to disseminate the information. If they
use it prematurely, it won't work. The grace period I recommend to UUCP
sites is "wait a week or so after the information appears in comp.mail.maps",
which on the average is 3 weeks after approval. As with any other major
change, you should anticipate startup problems and latency.
# What is so evil about listing the root server in the maps as an
# absolute worst case gateway to .CA ? If the subdomain information
# is as up to date as you claim, the path will hardly be used anyway :-)
A friend of mine put it very succinctly: If a generic .CA route exists,
then there will be no incentive to update the local maps instead of relying
on the .CA gateway(s). In short order the traffic pattern *will* change.
Even if the .CA gateway(s) is willing to handle more traffic, its neighbours'
facilities will be used without having a say in the matter.
rayan