& cplSiteName &
Comments
Newest First  |  Oldest First  |  Threaded View        ADD A COMMENT
<<   <   Page 2 / 3   >   >>
nemozen
nemozen
12/5/2012 | 2:57:19 AM
re: Internet Peering on Thin Ice?
LINX, MAE, PAIX, etc. are the physical points where peering (settlement free interconnects) takes place. These are typically ethernet switch fabrics, where each ISP need only get one port in order to exchange traffic with all the others that are present there. This is especially attractive if you need to peer with many ISPs in one location.

Most of the largest tier 1 peering interconnections are done on private circuits though, not thorugh these public peering points, since the traffic volume justifies the cost of a dedicated link between the two networks.

-nemo-
Tony Li
Tony Li
12/5/2012 | 2:57:19 AM
re: Internet Peering on Thin Ice?

In general, the IXs also support private peering, partial peering, and paid peering as well as SFI.

The big players tend to shift traffic away from the IXs when possible, just because a circuit to an IX generally costs more than half a circuit to your peer and you save a few router ports.

Tony
fiber_r_us
fiber_r_us
12/5/2012 | 2:57:18 AM
re: Internet Peering on Thin Ice?
Also, the IXs can not really support the volume of traffic that go between even two of the big players. I hadn't looked recently and don't know the exact numbers, but the majors (Sprint, Level3, AT&T, MCI, etc) SFI peer independedntly with each other in about 10 locations around the country with each location being a 10G connection.

That is, Level 3 has:
~ten 10G connections to AT&T, *AND* has
~ten 10G connections to MCI, *AND* has
~ten 10G connections to Sprint, etc...

Thus, it is possible that 10s (or even 100s) of Gigabits/sec of traffic can flow between two major backbones.

The big backbones are also connected at the "public peering" points, primarily for connections to the dozens (hundreds?) of smaller ISPs who simply don't require a lot of peering bandwidth in a lot of locations.

As Tony mentioned, peering at these Internet Exchanges (IXs) can be any of the forms of peering that I described before (SFI, paid peer, paid transit, etc). Each company makes its agreement with each other company independently.
vrparente
vrparente
12/5/2012 | 2:57:15 AM
re: Internet Peering on Thin Ice?
How can L3 be Tier one if they don't carry Cogent's routes ? Regardless of how they get them L3 can hardly claim to be Tier I if they don't carry them.
fiber_r_us
fiber_r_us
12/5/2012 | 2:57:14 AM
re: Internet Peering on Thin Ice?
No provider carriers "all" routes. In fact, it would be difficult to define what "all" even really means. The Internet is not completely "contiguous" from a routing perspective. There are always holes that are not reachable for various reasons.

Now, in the case of Level 3 not being able to reach Cogent, this is Cogent's problem for not arranging either paid peering with Level 3 or transit through Verio (or some other third party) for connectivity to Level 3.

Level 3's Tier 1 status is derived from their network customer base and their SFI interconnects with other Tier 1s. Connectivity between Level 3 and Cogent (a non-Tier 1) is Cogent's responsibility and has nothing to do with Level 3's "status".

Why should Level 3 (or any other provider) be responsible for Cogent not wanting to pay for traffic that Cogent wants to dump on them in order to maintain their cheap rates??? Either Cogent *pays* for the traffic, or they don't get connected.
rjmcmahon
rjmcmahon
12/5/2012 | 2:57:13 AM
re: Internet Peering on Thin Ice?
What are Cogent's reasons for not raising rates?
fiber_r_us
fiber_r_us
12/5/2012 | 2:57:12 AM
re: Internet Peering on Thin Ice?
Because their business model was to be the "low cost" provider to "business customers" (i.e. servers: connections that *generate* lots of traffic, but don't *consume* much). The idea being that you can sell Internet connections at $10/Mb/mo if your infrastructure only included this specific, easy-to-serve customer base.

However, this ignores the reality that paid peering/transit to other providers costs at least twice that much! So, Cogent's model only works if you can convince other providers to carry your *server* traffic to the *consumers* on the other provider's networks at little or no cost to Cogent.

Other providers actually have been carrying Cogents traffic for free since the bubble. Then, in the last few years, AOL, Sprint, a few others, and now Level3 have called bullsh*t on the whole plan and refused to continue to carry Cogent's traffic for free. Go figure, I'm amazed that Cogent has got away with it this long.
PO
PO
12/5/2012 | 2:57:07 AM
re: Internet Peering on Thin Ice?
While I've appreciated the explanations which have been given, I thought a couple other comments might broaden the subject a little bit more, using more colloquial and less formal terms.

First, the concept of a "stub" AS (Autonomous System), at least as far as Internet Routing is concerned. This would be a network which does not provide any transit services for other networks. It might be a corporate network, or it could be a regional ISP (even one that provides Internet access for some corporate customers which themselves might operate "stub" AS's ... but let's not get confused by the reality).

Then, we have all heard of the concept of "hot potato" routing: that a network will hand off traffic to the next network in the chain towards the destination AS at the earliest opportunity.

Peering agreements, of course, are what governs what that "earliest opportunity" actually is: they limit the advertisement of what ASs you can reach through any particular other network.

Even at the same Network Access Point (NAP, or Point of Presence (PoP); MCI's are known as MAEs - they say it's not an acronym although it supposedly originally came from "Metropolitan Area Ethernet" back when they were regional peering points ... but I digress) different carriers with a presence might have different commercial arrangements, different interconnect rules (google: Routing Arbiter), and therefore very different BGP tables and different reachability. It's not just a "free-for-all" of interconnections.

There was a time when MCI (mae.net), Merit (merit.edu), NANOG (nanog.org), and ISI (e.g. http://www.isi.edu/div7/) had much more useful information available online; however, I can only find a shadow of it now. And I've already demonstrated to those in the know just how long I've been away from the current details.

Since Google brought it up, I'll also refer to sbcbackbone.net and to wikipedia.org/wiki/PAIX (which tells me I have to re-learn some more acronyms which have changed meaning ... sigh).

I hope that folks will understand that Internet Peering is competitive, even if the details can become rather complex, and isn't on any thinner ice than other parts of the economy.
PO
PO
12/5/2012 | 2:57:07 AM
re: Internet Peering on Thin Ice?
Note to self: never try to guess what any given message board does with Anchor tags and html.

When I posted, the links were mucked up. In crafting this "reply", they all look fine. Not sure which form they'll take tomorrow, here they are again:

http://www.mae.net/
http://merit.edu/
http://nanog.org/
http://www.isi.edu/div7/
http://sbcbackbone.net/
http://wikipedia.org/wiki/PAIX
nemozen
nemozen
12/5/2012 | 2:56:59 AM
re: Internet Peering on Thin Ice?
> I hope that folks will understand that Internet > Peering is competitive, even if the details can > become rather complex, and isn't on any thinner > ice than other parts of the economy.

Peering is competitive, yes, but that does not mean it's not on thin ice. In fact peering (I mean SFI) is an unstable equilibrium state, because there's a discontinuity between peering (SFI) and the alternatives (paid peering or transit).

It's hard for a relationship to evolve from peering to non-peering without a huge disruption on the cost side. When Cogent gets de-peered, the price of the same Mbps to the same destination on L3 jumps from $0 to the transit price of the provider that Cogent could use to reach L3 (*).

Once the traffic ratios, or the business relationships, or any aspect of the peering justifications change, it becomes like two tectonic plates pushing against each other. The tension builds and builds, until it is released in an earthquake, with unpredictable damaging effects all around.

Cogent got caught with their pants down because the prices they charge their customers are artificially low, and locked in with flat-rate contracts, so Cogent cannot cover their true costs which include a certain probability of having to pay for transit to Level 3.

Now if both sides were interconnecting on a real market-price basis, they would always have a price at which they can exchange traffic. Sometimes it would be equal value in both directions (as in SFI), so the net flow of money would be $0. But, other times, the value in one direction would be more (either because there's more traffic or because its a higher value network), and the market prices would adjust to reflect that fact.
http://www.invisiblehand.net

-nemo-



(*) I mean the IP bandwidth cost -- of course there are other costs e.g. ports, circuits, etc. which are roughly unchanged between peering and transit.
<<   <   Page 2 / 3   >   >>


Featured Video
Upcoming Live Events
October 22, 2019, Los Angeles, CA
November 5, 2019, London, England
November 7, 2019, London, UK
November 14, 2019, Maritim Hotel, Berlin
December 3-5, 2019, Vienna, Austria
December 3, 2019, New York, New York
March 16-18, 2020, Embassy Suites, Denver, Colorado
May 18-20, 2020, Irving Convention Center, Dallas, TX
All Upcoming Live Events