Peering moves down to the Ethernet layer

Richard Chirgwin explains Internet peering and how it is changing to use more of the IP stack.

Although ISPs routinely use Ethernet interfaces to interconnect their routers in a peering exchange, the hard work usually happens up at the IP layer where routes are decided.

Alongside the traditional peering model, however, Ethernet is becoming the basis for traffic peering in a growing number of exchanges. With Equinix expecting to bring its Ethernet peering services to Australia within the year (along with 13 other countries), SearchNetworking spoke to the senior product manager Craig Waldrop from Equinix’s Denver offices.

In general, ISPs peer for a couple of reasons: one is to reduce their upstream traffic charges by passing traffic directly to each other. Internode and iiNet, for example, might find that there’s enough traffic travelling between their subscribers to justify setting up their routers so that this traffic doesn’t traverse their transit links. If they’re both present in the same peering exchange, all they need to do is make the connection between their routers, and configure their routes to use that connection.

Apart from cost savings, ISPs might establish a peering connection for performance reasons – for example, to reduce the latency to a popular games server, thus making its service more attractive to the games community.

This traffic-passing is handled by routers – iiNet configures its routers so that Internode traffic passes over a particular interface that’s connected to Internode’s routers, and vice-versa.

Peering at the IP level is well-established and well-understood, so SearchNetworking asked Waldrop what kinds of customers and applications created a market for Ethernet-based peering.

“The key difference is that this service is offering peering for private networking purposes rather than public,” Waldrop said. “Instead of [the peers] exchanging routes to the Internet, we’re passing off an Ethernet frame specific to a customer, and tagged at Layer 2.”

The point of the exchange

The idea, he said, is for customers such as carriers to simplify their interconnect for enterprise data traffic.

A carrier serving a business with offices in Sydney and California will need a partnership to service one city or the other. Typically, the carriers will exchange that customer’s traffic over a “vanilla” TDM or SDH-based data link – and they may have to do a lot of work to make sure that they’re offering the “same” Ethernet service (in terms of service levels, traffic prioritisation and so on) at each end of the link.

Equinix’s hope, Waldrop said, is that by letting carriers exchange customer traffic as Ethernet, they will be able to offer a more seamless end-to-end experience.

The Ethernet exchange, he said, “is about carriers extending their footprint.

“Telstra has local access in Australia, but if it needs to facilitate private Layer 2 Ethernet services to someone else’s network, they could do that through a private connection, or in an exchange.”

By creating exchanges in which Ethernet traffic can be handed off, he said, the exchange of traffic becomes more scaleable. The carrier gets access to a number of additional partners, giving it advantages both as buyer and seller. As buyer, it can extend its network coverage across the networks of those partners; and as seller, it can attract other carriers to reselling services across its network.

“As a carrier, I can connect once to multiple buyers and, at the same time, extend my own services across multiple partners,” he said.

And carriers participating in the exchange also get more flexibility in their interfaces. Instead of having to choose between the standard inter-carrier traffic increments, they can pass traffic to each other at whatever rate they choose to offer.

The challenges

The key to the success of the concept, Waldrop said, is carriers’ willingness and ability to align their Ethernet service offerings.

“What we’re going to have to facilitate is a brokering of a lot of information about [participants’] Ethernet services,” he said.

For each participant, he said, Equinix collects information about the Ethernet service offerings, so that “carriers can partner those who are best qualified to meet their needs.”

He said carriers also demand redundancy, both of the switches in the exchange and across multiple data centres. “That’s how the carriers design their networks for protection and diversity.”

Most important, however, is how well different services map to each other. “We’re trying to make that easier by creating a template and a service profile that maps different carriers’ Metro Ethernet services. We look at things like class of service, supported Ethernet frame types, and the template allows those services to map from provider to provider.”

And what would all this offer for the end user?

“Customers are looking for a consistent service experience whether the office is in Sydney or Denver or London. They want the service to operate the same, and in the Ethernet exchange, we can facilitate that.”

Read more on WAN performance and optimisation