Here are my first thoughts on a tiered model, with attributes of each tier. Top level (Tier 1): -- provide transit as their principal business -- have at least some default-free routers -- have connectivity at >1 geographically separated major exchange -- need special measures to deal with BGP scaling issues inside their AS (or multiple AS) such as confederations, clusters, etc. 2nd Level -- provide transit within a geographic area -- may have default-free routers -- limit operations to a geographic area; may connect to multiple exchanges within that area 3rd Level -- do not provide commercial transit services, although they may incidentally provide transit among their customers -- do not do BGP peering with any "downstream" organizations. Their user base is part of their AS. -- May peer with multiple upstream providers 4th Level -- do not run BGP -- Internet access through provider only
Is this "on" the Internet or is it "the" Internet? (btw, I still take exception to the term default-free. Taken at face value, its just about as credible as a prior poster who claimed that the tail end of his 28.8 modem was the internet backbone. Proxy Aggregation is presuming some level of default behaviour and needs to be considered as such.)
The only main difference that you have between Tier 1 and Tier 2 levels is the geographic sensitivity of the public exchange. This may be a false criteria or maybe a vindication of S.Deering. You also make zero distinction between public and private exchanges. --bill
I understand your point re: default routing, but: Default-free is still a useful distinction and is easy to see. If you are a customer of a provider, you can easily traceroute to a nonexistant IP address and see at what router it stops. Now, many 2nd level providers that *could* operate default-free choose not to. Even if you have three or more sets of 30k+ routes each, it takes balls to risk dropping packets that your customers want you to deliver just so that you can have the packet be dropped at your router instead of at your (possibly backup) transit provider's router.
--bill
Avi
On Mon, 8 Apr 1996, Avi Freedman wrote:
Can't anyone who takes full routes from any tier 1 provider operate without a default route? And isn't it a reasonable assumption that if you don't have a route somewhere, odds are they don't have a route to you (assuming you do your own BGP routing) and so a default route is mostly pointless anyway? What am I missing? DS
Sorry if this thread is taking too much space on NANOG. I didn't start it; also, it's been suggested to me that inet-access may be a better forum this. I'm not sure you're missing anything, but here are some things that we consider: a) It can take a minute or so, depending on link speed & router types, to load 30k+ routes into a crisco. Therefore, if you have no default route, even though your provider may have routes/connectivity to sites, you may drop packets in the meantime. b) Your provider may in fact have you statically routed even if they speak BGP to you, you'd have to check - esp. if you're singly-homed. Ever since mid-last year, the consensus was that localized flapping of connections shouldn't be reflected in all of the routers on the 'net if at all possible. (see appropriate pointers on route dampening and inserting Null0 routes) c) It takes balls to risk dropping customer packets on the bet that just because your provider doesn't have or hasn't told you a route to a destination, they can't get there. It's a good bet, but not a sure one. Avi
Because not all tier 1 providers have all routes. In fact, most of them don't. They get "full" routes by peering with many others or proxy aggregating and then trying to do the "right" thing. It really depends on who/where you want to reach. -- --bill
David, I think what Avi meant was that if you take full routing table and not using default, chances are that no matter how many backup providers you have, you are still risking dropping packets on *YOUR* router if routes to a particular location are lost. This makes you look bad when your customer traceroutes. If you have default then even if all the external routes get lost you can still deliver the packets to your provider and let him drop it. This looks favorable from your customer's traceroutes. In practice, it is better to drop it as early as possible because any additional delivery is unnecessary, but... Jun -- o o o o o o . . . ___========_T__ ___========================_T__ o _____ || Jun J Wu | | jun@wolfox.gsl.net | .][__n_n_|DD[ ====____ | Global One | | http://wolfox.gsl.net/jun | |
(________|__|_[________]_|_____________|__|____________________________|_| __/oo OOOOO oo` ooo ooo 'o^o o^o` 'o^o o^o`
===== David ``Joel Katz'' Schwartz previously wrote: ====
David, You're right you don't need a default route. However, imagine that you don't have confidence in the routing table entries. Imagine a situation like this: NAPX---NSP1a-------NSP1b-----NSP1c-----NSP1d-----NSP1e-----+ You Imagine that you take full routing tables from NSP1e. There is a bit of latency in him getting a route propogated, so conceivably with some flapping issues, you might not have it if you wanted it. So, what you could do is default to NSP1e. If you're multihomed, as such: NAPX---NSP1a-------NSP1b-----NSP1c-----NSP1d-----NSP1e-----+ | You +-----NSP2a-------NSP2b-----------------------------------+ Concievably you could have wacky problems at 'you' which could cause you to drop packets. By having a default, you entrusted the delivery in those situations to one of your upstreams. -alan ps. we multihome to 2 providers, we take full routing tables, and we don't have an external default. I'm not recommending the above logic, just expounding on a possible explanation..... ......... David ``Joel Katz'' Schwartz is rumored to have said: ] ] On Mon, 8 Apr 1996, Avi Freedman wrote: ] ] > Now, many 2nd level providers that *could* operate default-free choose ] > not to. Even if you have three or more sets of 30k+ routes each, it ] > takes balls to risk dropping packets that your customers want you to ] > deliver just so that you can have the packet be dropped at your router ] > instead of at your (possibly backup) transit provider's router. ] > ] > Avi ] ] Can't anyone who takes full routes from any tier 1 provider ] operate without a default route? And isn't it a reasonable assumption ] that if you don't have a route somewhere, odds are they don't have a ] route to you (assuming you do your own BGP routing) and so a default ] route is mostly pointless anyway? ] ] What am I missing? ] ] DS ] ] ]
On Mon, 8 Apr 1996, David ``Joel Katz'' Schwartz wrote:
Yes
that if you don't have a route somewhere, odds are they don't have a route to you (assuming you do your own BGP routing) and so a default
Yes/No
route is mostly pointless anyway?
Ok, why look through 30K routes instad of just sending all packets out the default.
What am I missing?
A lot. Nathan Stratton CEO, NetRail, Inc. Tracking the future today! --------------------------------------------------------------------------- Phone (703)524-4800 NetRail, Inc. Fax (703)534-5033 2007 N. 15 St. Suite 5 Email sales@netrail.net Arlington, Va. 22201 WWW http://www.netrail.net/ Access: (703) 524-4802 guest --------------------------------------------------------------------------- "Therefore do not worry about tomorrow, for tomorrow will worry about itself. Each day has enough trouble of its own." Matthew 6:34
Some nits. In message <v01520d0bad8c60bc8902@[168.143.1.215]>, Howard C. Berkowitz writes:
principal IP business (some people provide voice services or do other things).
-- have at least some default-free routers
Span at least one continent with default-free routers. Are able to provide customers with default free routing as an option.
-- have connectivity at >1 geographically separated major exchange
Yes. And span these with default free routers (ie: taking full routing at two routers and defaulting in the middle doesn't count).
-- need special measures to deal with BGP scaling issues inside their AS (or multiple AS) such as confederations, clusters, etc.
or multiple AS to deal with scaling... but if a really hefty router comes along or we ever are able inject AS paths into the IGP, this no longer applies. How about connection speed? At least a DS3 backbone? A redundant backbone (no single circuit failure can partition the provider)? Can we squeeze in 1.5: -- provide transit as their principal business -- have at least some default-free routers -- have connectivity at >1 geographically separated major exchange -- do not have a default free backbone -- maybe cannot provide full routing to all but a few customers -- maybe not redundant DS3s Nobody in particular in mind here.
Connection speed? Is Nearnet (mostly New England) circa 1994 equivalent to some small provider with two T1s and a bunch of routers in a limited geographic region? Again, having an providing full routing is a big factor too.
participants (8)
-
Alan Hannan -
Avi Freedman -
bmanning＠isi.edu -
Curtis Villamizar -
David ``Joel Katz'' Schwartz -
hcb＠clark.net -
Jun John Wu -
Nathan Stratton