MPLS sets priorities for Internet data flow

Multiprotocol Label Switching (MPLS) is on the verge of adoption as a standard, but that will not settle debates about business, technology and philosophy that the protocol raises.

MPLS may be the potion that transforms the public Internet from a best-effort network not good for mission-critical applications to a network that could meet the most stringent business requirements.

But so far, MPLS has not delivered on its promise. Operators and vendors say it is too early to expect significant results, since only a few carriers have implemented the technology and few vendors have standardized on it. MPLS today is just a specification, though the Internet Engineering Task Force is believed to be close to ratifying it as a standard.

In the circa-2000 Internet, MPLS exists as an overlay technology offering the owners better tools for managing their Internet Protocol (IP) networks and their customers' access to premium services.

What network operators do next may define the future of the technology and, to some extent, the future of the Internet. Companies will have to decide between investing in a common infrastructure with no immediate returns--Internet-wide deployment of MPLS and peering MPLS traffic--or investing in a particular project with a clearly defined goal--building a for-profit MPLS overlay network. Either option could terribly slow the Net's development for the next 30 years, or boost the medium's potential to new heights, depending on who is talking.

The basic principle is that each IP packet is identified according to the type of information it carries - a telephone conversation, a videoconferencing bridge or a rush e-mail message. Based on this, identification routers "know" which traffic has priority and regulate the traffic flow accordingly.

It may be possible to deploy MPLS across the entire fabric of the Internet because all Web traffic is routed. Setting up identification rules as well is just around the corner, since MPLS is about to become a standard. Once the standard is established, all devices running the protocol would know what each packet identification stands for.

Since MPLS is a labeling protocol, not a routing protocol like Border Gateway Protocol, it can work over any kind of data network. It could provide much needed end-to-end continuity for the Internet as it rides over networks that are modern or antiquated, slow or fast. But that scenario can only play out if network operators agree to include MPLS in their peering arrangements.

By engineering peering exchanges to support the labeling protocol, network operators could answer users' pleas for application-specific, dependable performance. Today, carriers peer their networks, or barter long-distance traffic, based on volume.

The introduction of MPLS suggests carriers would have to begin exchanging videoconferencing traffic for videoconferencing traffic, voice for voice or perhaps renegotiate deals based on which company is sending best-effort traffic and which is sending "premium" traffic.

Since the IETF has not addressed MPLS peering on the industry-wide level, many engineers say the need to establish such a setup is a no-brainer.

"I am an engineer, not a politician or a financial or sales- person. Maybe I live in a naive environment amongst engineers, where we assume that the global Internet is good for everyone," says Steve Carter, Global Crossing's director of engineering and operations.

Global Crossing has implemented MPLS throughout its network, which runs on Cisco and Juniper Networks routers, and is looking into adding the feature on Unisphere Networks equipment. "Maybe the Internet has changed. Maybe the Internet now is not like that. Maybe it is now controlled by bean counters who want to make sure they get their value for the money - I don't know," Carter says.

The decision is not so obvious for engineers who have already been burnt trying to make the Internet a better place by developing better peering exchanges. This crowd has a more capitalistic view of the matter, contending that so long as there is no financial incentive for networks to perform better - with or without MPLS - quality of service (QOS) will never grace the Internet with its presence.

CoreExpress, for example, is a service provider that specializes deploying QOS-type technologies in order to become a carrier of prioritized traffic.

Its founder, Michael Gaddis, earlier tried to unite mid-sized Internet service providers (ISPs) into a cartel that would peer with the Internet's largest backbones as an equal, therefore improving economies of their operations and delivering faster Internet service to end users.

The project died a slow death. Gaddis' new venture is based on a firm belief that for-profit businesses are deeply egoistic, and therefore have to attend to their bottom line first and a better life for the underprivileged competitors second.

"The cost that ISPs are going to incur to deploy QOS are only going to pay off if they charge a customer premium to offset that cost," says Tony Zeis, chief technical officer at CoreExpress. "While it is our hope and expectation that all of the networks adapt quality of service, we do see a problem for ISPs attempting to use their pre-existing peering contracts in attempting for this quality of service to occur."

CoreExpress' service - which in essence provides QOS on demand using MPLS and other protocols such as Differentiated Services and Type of Service - is currently in beta tests. And Global Crossing's Carter still doesn't see, from a technical standpoint, how MPLS-based peering could be done. So both companies must continue arguing their positions. Meanwhile, MPLS is catching on as an intra-network technology.

MPLS is good for making IP traffic move more gracefully over Asynchronous Transfer Mode (ATM) fabric, scaling the network by making it more efficient and creating virtual private networks (VPNs), according to Cisco's senior director of marketing, Robert Redford.

In other words, individual carriers use MPLS to either engineer their networks to perform identically despite using different routing technologies, or to develop QOS-based services. But, again, these advances have been made only inside individual networks.

Global Crossing, which spent about a year implementing the technology, doesn't need to use MPLS to overcome ATM routing efficiencies since its network is 100 percent IP. It does use the protocol for scaling and to create VPNs.

"Our network would not operate without MPLS in place," Carter says.

Having acquired a number of networks and replaced an ATM backbone with packet-over-Synchronous Optical Networking technology, Global Crossing was forced to purchase circuits from vendors that didn't meet its physical goals as it rolled its holdings into one network. By deploying MPLS and separating routing and forwarding protocols, the carrier has been able to extract needed bandwidth from circuits that wouldn't have the juice to carry Global Crossing's traffic if they were engineered with a routing scheme other than MPLS.

The next stop for Global Crossing is to develop a VPN service based on the protocol.

These kinds of tactical developments are of paramount importance. As they see more carriers developing services based on the technology, vendors such as Unisphere are encouraged to add MPLS to their routing products.