Ethernet Frame Relay

I’ve been talking to a lot of people about optical Ethernet lately, trying to determine just what it will take to create a profitable carrier service around this “new new thing.” Along the way, I met Roy Bynum, formerly of WorldCom Inc. (Nasdaq: WCOM), now an independent consultant working out of Texas.

Roy was the instigator of the 10-GigE WAN PHY, as well as a co-author of the X.86 LAP-S Ethernet-over-Sonet protocol and the X.86 Ethernet Flow Control Specification. He also was the architect of what is now commonly referred to as the Ethernet Private Line, which is, by most accounts, the only truly successful (read: "profitable") Ethernet-based service from a carrier today.

Roy is a big advocate of Ethernet. He sees it as not only the “next big thing,” but as the likely successor to the throne currently occupied by Frame Relay in the data networks of service providers – and by T3 on the transmission side of the house. This is big. It means Ethernet ultimately emerges as the winner in both transport and switching, just as it has within private enterprise networks.

Roy’s history lesson goes this way: A little less than a decade ago, Frame Relay overtook X.25 because it was cheaper for the service provider to operate and cheaper for the customer to own. Frame Relay could take a user up to 1.5 Mbit/s easily, which was big bandwidth at the time. Frame Relay service facilities are much more scaleable than X.25, thanks to encapsulation of customer frames in Asynchronous Transfer Mode (ATM) or Internet Protocol (IP).

That virtual switching fabric also allowed the service providers to distribute the Frame Relay access gateway facilities, which meant that the Frame Relay customer access did not need to be backhauled to large centralized switching points, making the local loop access a lot less expensive.

It also made it easier to build a redundant, diversely routed network architecture to support the interconnections among Frame Relay gateway facilities in different parts of the country. As history proved, Frame Relay was actually a better service, a better product, with overall better cost of ownership for both the service providers and the customers. Frame Relay opened up new markets for packet services by providing long-haul point-to-point and point-to-multipoint connectivity without the distance charges of private line or the volume charges of X.25.

For a while, people thought ATM would supplant Frame Relay once bandwidth demands from end users grew, but it didn’t: too expensive, too complex, violating the rules of low cost of ownership for carrier and customer alike. But along comes Ethernet, which today is appearing as an interface on just about every piece of data and transport networking gear in the service provider, after completely overtaking the enterprise.

Where Ethernet has already started to make an impact through simple Ethernet private lines (Ethernet mapped directly to Sonet channels), it’s worth taking some time looking at how Ethernet may just take over the packet services market as well. The form it takes may not be so different from Frame Relay, just faster and cheaper, and since Ethernet is based on its own set of standardized frames, it may make sense to just call this new service and infrastructure Ethernet Frame Relay.

The key, Roy argues, is Multiprotocol Label Switching (MPLS) encapsulation of Ethernet. No funny business with customer frames is allowed, giving users the utmost sense of security and giving carriers a way to carry that traffic without having to process it at anything other than Layer 2. Since the introduction of Frame Relay there have been several changes to the Internet Protocol standard that allows Layer 2 switching of end-to-end flows. This, in effect, creates a distributed switching fabric at the data-link layer, where before it was at the network layer. This should provide a much more stable and cost-effective way to implement the same type of encapsulated frame switching that was done for Frame Relay ten years ago.

Ethernet, the most commonly used protocol in the world, has matured in the last ten years. Legacy-free service providers experimented with providing direct Ethernet VPN (virtual private network) access to Internet services using Gigabit Ethernet as the transport protocol. Ten-Gig Ethernet has added a Sonet-like overhead to support managed optical wavelength services. The Institute of Electrical and Electronics Engineers Inc. (IEEE) P802.3ah Ethernet in the First Mile (EFM) Task Force is in the process of defining subscription network service-provider management support. Ethernet, instead of being only for privately owned enterprise networks, is now becoming a basis for building new types of revenue-generating data transmission services.

The advent of MPLS and EFM at the same time now makes it possible to provide the same kinds of networking services for Ethernet that were previously provided using Frame Relay. By using MPLS to encapsulate Ethernet frames, a distributed virtual switching fabric can be built for long-haul Ethernet point-to-point and point-to-multipoint links without having to pay long-distance private line charges. In addition, Ethernet/MPLS gateway switches can be distributed throughout metropolitan areas in large as well as small cities, making the virtual switching fabric even more distributed. This is exactly what is needed to create the next generation of wideband data transmission services.

MPLS has better QOS (quality of service) and bandwidth reservation capability than the older Internet Protocol that was used for Frame Relay. In addition to the ability to provide tiered bandwidth management to provide the same PIR/CIR (peak information rate over committed information rate) relationship, the use of Ethernet active flow control in the local access loop makes it possible to provide a more granular PIR as well.

Additionally, active flow control might be used to control the CIR/PIR relationship by using the MPLS forward and backward congestion control to generate a flow control message to the customer’s Ethernet switches on the ends of the access links. The active flow control message would slow the customer's transmission of Ethernet frames down to a level that prevents too much data loss during times of congestion. This will be a big help during peak business hours or at times when a fault in the transmission network reduces the service capacity of the distributed switching fabric.

With this, Roy Bynum believes, broadband access technology can be used to provide a wideband PIR for the narrowband CIR. Over a period of time, given the right market circumstances, it will provide an evolutionary migration path from wideband/narrowband packet service demand to broadband/wideband packet service demand and beyond. Where the need exists to use the existing copper local access infrastructure, 802.3ah is also defining an xDSL-based form of subscription network managed Ethernet. This wide variety of physical access standards will support the existing Frame Relay narrowband access as well as the growth to the high bandwidth of physical access.

See? That’s big. For enterprise customers without fiber ot the building, Ethernet-over-copper solutions get them on the network via a simple LAN interface. For those with access to fiber on metro optical rings, Ethernet access is already possible via Ethernet-over-Sonet or dedicated GigE links to service provider nets. Adding this encapsulation of Ethernet via MPLS is the final ingredient to taking Ethernet from an access service to a switched network service, where the real money is. Carriers haven’t ever been able to generate a lot of profits from packet services, just revenue. Something needed to come along and truly disrupt the cost of ownership of packet services to really get carriers serious about deploying broadband data services infrastructure. That means carrier spending, and that’s good for everyone.

— Scott Clavenna, Director of Research, Light Reading

Sign In