It is amazing how a subtle improvement in technology can lead to a whole new industry. Downhill skiing was cool in the 1960s but became a bit “passé” in the late 70s and early 80s. Then, in the late 80s and 90s, the snowboard burst onto the scene and jolted the somewhat stodgy ski culture into a whole new market for “Generation X”.

A similarly subtle yet revolutionary change is in the process of happening to Ethernet, creating a brand new and very exciting industry around the concept of “Carrier Ethernet Transport” or CET. CET offers a new option for the transport of Ethernet services in carrier networks.

To understand the substance and significance of the CET “snowboard”, it is first important to understand where so-called Next Generation Network (NGN) thinking has been and how it is changing.

Evolution to Next Generation Networks

Over the last ten years or so, the rate of evolution for Ethernet services infrastructure has been dramatic.

Not that long ago, enterprises would either lease SONET/SDH circuits between routers or buy Frame Relay services. While there is still a significant installed base of enterprises using these technologies, native Ethernet network interfaces on enterprise routers and Ethernet switches have really started driving a huge demand for the transport of native Ethernet services in carrier networks.

Saddled with SONET/SDH transport networks, carriers tried a few approaches to stuff the square peg of Ethernet into the round hole of legacy infrastructure. Dedicated builds (e.g. Fast Ethernet at 100 Mb/s or GigE on dedicated fibre, with or without WDM), RPR, and VCAT have all been tried, in some cases with great success in terms of market share and revenue. However, that success brought scalability issues, inefficient use of fibre assets, high cost for the delivery of Ethernet services (without the growth in profitability that one would expect from economies of scale), and limited Operations, Administration and Management (OAM) capabilities and Service Level Agreement (SLA) options.

Carriers have recently begun replacing their SONET/SDH and PSTN networks with Next Generation Networks, moving telephony to VoIP and soft-switching, and using an IP/MPLS core for all services: voice, broadband, Ethernet and, in some cases, video, gaming, and mobile traffic as well.

In parallel to this, Ethernet has become a universal protocol for enterprise services, wholesale services (e.g. GigE interfaces on IP DSLAMs), and applications into the home. By some estimates, 95% of carrier traffic originates as Ethernet, making Ethernet a logical choice for not just service interfaces, but also, where possible, transport infrastructure as well. The volume of Ethernet systems and components deployed around the world has driven the price points of Ethernet to the level where it can and must be considered as an option for carrier transport networks, not just client interfaces to customers.

Today’s Choices for Ethernet Service Infrastructure

The main question now facing carriers who either have moved or are moving to IP/MPLS-based NGNs is this: given the low cost of Ethernet systems and components, to what extent can and should Ethernet be used in the carrier transport network?

Before CET, two main approaches developed for next generation carrier transport network design to support Ethernet services as well IP services:
Stand-alone IP over WDM: the transport of Ethernet services as IP traffic, as part of a large IP/MPLS network, with IP/MPLS routers at all points in the network connected by “dumb” optical bandwidth.
Carrier Ethernet Overlay: deployment of an intelligent Ethernet transport and switching layer between layer 3 MPLS and the optical transport layer, to address cost and scalability issues of IP over WDM.

IP over WDM

The IP over WDM solution introduced desirable connection-oriented MPLS Quality of Service (QoS) to IP services.

However, carriers have found this approach expensive to deploy to the edge of their network (i.e. MPLS routers at each access point to the network edge).

Carriers have also encountered operational complexities and scalability issues in managing the large number of MPLS LSPs that result from this architecture.

One of the most serious problems with this architecture is how it consumes network resources for end-to-end Ethernet services. As shown in Figure 2, a GigE or sub GigE service “trombones” from the optical layer to the MPLS layer and back down again at each network hop, until it reaches its intended destination.

This uses the capacity of MPLS routers at each hop (optical ports, processor/table/memory capacity) for traffic that could otherwise bypass MPLS if an optical transport mechanism were available to do so.

By some estimates, 70% of all carrier IP traffic is transit traffic and yet is subjected to “tromboning” routing.

Carrier Ethernet Overlay

This approach improved on the “MPLS everywhere” design of IP over WDM by introducing lower cost VLAN-based Ethernet aggregation and switching between the optical transport and MPLS layers, and the ability of VLAN switching/swapping/stacking technology to add more scalability to MPLS.

The use of lower cost Ethernet components reduces the number of more expensive MPLS resources in the network, and leads to an overall reduction in network cost compared to IP over WDM.

However, it also introduces some undesirable aspects of Ethernet:
Connectionless aspects of Ethernet are not required for carrier Ethernet transport (e.g. PBB-TE), it unnecessarily consumes optical bandwidth and isn't scaleable (e.g. spanning tree, broadcast unknown, MAC learning).
Ethernet also lacks QoS and OAM.

In general, while this approach solves some cost and scalability issues, it adds an additional layer to the next generation IP/MPLS carrier network. Furthermore, rather than addressing the MPLS/optical “tromboning” of the IP over WDM approach for end-to-end Ethernet services, it only moves it from the MPLS layer to the Ethernet layer.

The “Snowboard” Arrives: Carrier Ethernet Transport (CET)

Carrier Ethernet Transport offers significant CAPEX and OPEX improvements over these first two approaches for NGN transport of Ethernet services.

By combining emerging Ethernet Tunnel Switching technologies (e.g. PBT/PBB-TE and T-MPLS) with an agile optical transport layer that provides Wavelength Transmission, Wavelength Switching and Sub Wavelength Switching -- thus switching as much Ethernet traffic optically as possible -- CET:

• Reduces the load on expensive MPLS systems and precious LSPs, and
• Reduces the number of expensive optical ports required at metro/core hub points.

CET takes a holistic approach to next generation network transport:

• It recognizes the valuable role that agile layer 1-based Wavelength Switching and Sub Wavelength Switching in the optical transport network can play in solving the tromboning problem for Ethernet services, and optimizes (but does not replace) the use of the network’s MPLS resources.
• It introduces an Ethernet Tunnel Switching layer based on new technologies: specifically, PBT/PBB-TE and T-MPLS. This recognizes the valuable role that Ethernet and VLAN technology plays in optimizing MPLS cost and scalability, but addresses the weaknesses that Ethernet previously had as a transport technology (lack of OAM and QoS, LAN-oriented broadcast learning and routing, etc.).
• Rather than adding layers of complexity to the network (as “Carrier Ethernet Overlay” does), it collapses the network to two main layers: the MPLS service layer (implemented by IP/MPLS router vendors) and the transport layer (combining Wavelength Transmission/Switching and Sub Wavelength Switching, with PBT/PBB-TE and/or T-MPLS enabled Ethernet Tunnel Switching) into one network element.

Having a clear separation between the service layer and transport layer ensures the optimal scalability.

As shown in Figure 5 above, CET solves the “tromboning” problem for Ethernet services, optimizing the utilization of network components. Transit traffic is transported in the optical layer for as long as possible, and is only delivered up to the MPLS service layer when required. The result is fewer optical ports on MPLS routers and optical transmission systems, all of which are used with greater efficiency, and lighter load on precious and expensive MPLS resources (processor, memory, LSPs, etc.).

Other advantages of CET are that services can be guaranteed with SLAs, and that it optimizes the existing fibre plant.

The PBT vs. T-MPLS Debate

In recent months, debate has begun over which of PBT (or PBB-TE) and T-MPLS is the better technology for delivering the benefits of the Ethernet tunnel switching component of the Carrier Ethernet Transport architecture.

Both of these technologies fulfill the requirements and goals of tunnel switching:

• They both bring connection-oriented circuit capabilities to Ethernet, offering bi-directional tunnels with the same carrier-class transport features traditionally provided by SONET/SDH: QoS, protection and sub 50 ms protection switching via end-to-end OAM mechanisms.
• They both provide a smoother migration to connection-oriented Ethernet for carriers, allowing the possibility of re-use of the same planning, provisioning, operations, and management systems that have been used for the SONET/SDH network.

From a technology perspective, they both remove and add features to well-established technologies to make them more suitable for connection-oriented Ethernet transport. A summary:

PBB-TE:

Basis: PBB-TE (for Provider Backbone Bridge with Traffic Engineering) is under development by the IEEE using the foundations of PBB which are widely available in carrier layer 2 Ethernet switches today. Ethernet Provider Backbone Bridge (PBB) standards began some years ago and were approved in 2005. They extended the Ethernet transport and VLAN standards of the 1990s with almost all of the features required for connection-oriented Ethernet, including:

• The IEEE 802.1ah frame format used for PBB, built on mature IEEE 802.1q VLAN transport standards of the 1990s and including the carrier scalability and flexibility of MAC-in-MAC VLAN stacking and switching.
• The same frame forwarding mechanisms and addressing scalability provided by so-called Q-in-Q VLAN stacking and swapping according to IEEE 802.1ad, also already common in carrier layer 2 Ethernet switches.
• It also includes mature Ethernet QoS mechanisms (IEEE 802.1d, 802.1p, and 802.1q) also already widely deployed.

What it removes: It turns off some of the LAN-oriented features of Ethernet that impact Ethernet’s performance and scalability in a carrier transport network: specifically, Spanning Tree Protocol and MAC learning (flooding and broadcast of unknown MAC addresses).

What it adds: It adds OAM capabilities that enable SONET/SDH-like sub 50 ms protection switching: specifically IEEE 802.1ag Connectivity Fault Management (also worked through the ITU as Y.1731) that uses specific Ethernet frames to monitor end-to-end tunnel integrity.

T-MPLS:

Basis: Transport MPLS is under development by ITU Study Group 15. Its architecture was approved in early 2006, building on MPLS standards developed in the late 1990s. Standards for T-MPLS include G.8110.1, G.8101, G.8131, G.8121, G.8112 and define interfaces and protection switching, all in the same standard connection-oriented framework as SDH and OTN. ITU standards work continues on OAM, control plane (G.7718 and G.7715), and interworking.

What it removes: Functionality from MPLS that obstructs OAM so that SONET/SDH-like sub 50 ms protection switching can be achieved: specifically, Penultimate Hop Popping (PHP), Equal Cost Multipath (ECMP), and others.

What it adds: OAM capabilities that enable sub 50 ms protection switching: Y.1711 OAM and Y.1720 protection switching of LSPs.

So, what are the advantages and disadvantages of each approach? As they are both relatively new technologies, it remains to be seen if both will be deployed or if one will dominate.

• As of the writing of this article, a number of major incumbent carriers have publicly indicated their plans to deploy PBT, but there have not been any announcements concerning T-MPLS deployment.
• In general, PBT is expected to have economic advantages over T-MPLS. This is a result of its foundation in Ethernet technology and thanks to the high volume of Ethernet components and subsystems worldwide compared to MPLS.
• T-MPLS is more mature than PBT, having its basis in MPLS technology that has a longer history than PBB.
• Both are progressing at more or less the same rate in terms of standards refinement and the development of control plane standards.
• Both achieve the goals of CET tunnel switching, and provide a natural progression for carriers using SONET/SDH toward circuit-oriented Ethernet.
• For carriers who have not already made a widespread migration to MPLS, PBT as a technology is viewed as more familiar to their SONET/SDH infrastructure than IP/MPLS and T-MPLS.

Conclusion

It took a while for older skiers to accept snowboarding as a legitimate activity with which they could share a gloriously snow-covered mountain. Today, however, it is not unusual to see some of us forty-somethings “busting” down the hills on snowboards to keep up with our teenaged kids!

Similarly, IP/MPLS has certainly established itself as the common service layer for the NGN. However, the addition of complementary Carrier Ethernet Transport to it, focused on the transport layer, is ideal for transporting all forms of Ethernet traffic, including Ethernet services. Although it is a departure from pure MPLS, the MPLS world is already starting to appreciate what CET adds to transport.

I’ve even seen skiers and snowboarders sharing a half-pipe … so, it must be possible for MPLS and CET to share the same fibre pipe. … yikes, better stick with telecom.

Brian Pratt is Meriton’s Director of Technical Marketing, EMEA. He is based in Birmingham, U.K., where there are no mountains, and can be reached at brian.pratt@meriton.com.