Setting the Standards with Dr. Yaakov Stein

In this edition of “Setting the Standards,” rather than describe a particular meeting of a single SDO, I would like to talk about a subject that is being discussed in multiple SDOs, namely carrier-grade Ethernet. As more and more traffic is being delivered in Ethernet format, carriers are realizing the advantages to converging on a pure Ethernet infrastructure, but are not able to do so with the present Ethernet mechanisms. In a previous article I talked about the Y.1731 Recommendation developed by SG13 for Ethernet OAM, and this is an important step forward. ITU SG15 has also recently developed G.8031 for Ethernet protection switching, filling another lacuna. In a future article I will talk about methods for transporting synchronization over Ethernet networks, for those applications that need it.

However, the most critical problem lies elsewhere. Native Ethernet topology determination mechanisms are based on MAC learning, Spanning Tree Protocol (STP), flooding of unidentified frames, and aging of MAC addresses. As these mechanisms are not deterministic and may deliver frames to unintended termination points, they are unsuitable for carrier-grade transport networks. Carrier-grade networks can be based on Ethernet format frames, but require replacing MAC learning, STP, flooding and aging with deterministic setup of point-to-point connections. The setup may be accomplished via control-plane signaling protocols, such as GMPLS, or via management plane Operational Support Systems (OSSes), but must uniquely specify the path to be followed by frames associated with a connection, rather than allowing a flow to follow a path chosen by nondeterministic native Ethernet topology establishment mechanisms.

Three different mechanisms have been proposed for this purpose:

• PBT (Provider Backbone Transport), which has been proposed in the IETF as draft-fedyk-ethernet-gmpls-ethernet-pbt, and in the ITU as G.pbt

• PVT (Provider VLAN transport), also known as VLAN cross-connect, which in the IETF has been proposed as draft-sprecher-gels-ethernet-vlan-xc, and has been proposed in IEEE 802.1 as well

• T-MPLS (transport MPLS), presently described in ITU-T SG15 draft Recommendation G.8110.1.

All of these mechanisms endeavor to minimize changes to frame formats defined by IEEE 802.3 and to the existing Ethernet forwarding procedures. All support packet prioritization, QoS and use of traffic engineering, and can be easily augmented by protection mechanisms. All are presently defined for transport of conventional Ethernet clients, and require extension to carry generic client services.

PBT exploits the IEEE 802.1ah (MAC-in-MAC) format in order to achieve complete client/server separation. It uses the 60-bit address formed by combining the destination address and VLAN ID fields of the provider MAC as a globally unique backbone network edge identifier (this can be viewed as using VLAN IDs that are unique only with respect to a given MAC address). As the format also includes a unique source MAC address, OAM traceback is simplified as compared to less self-describing architectures. The tables of IVL Ethernet switches are populated by OSS configuration or GMPLS signaling rather than by MAC learning, thus forming a deterministic point-to-point connection. Once set up, Y.1731 OAM is used to maintain the connection. PBT can coexist with native Ethernet bridging by partitioning the VLAN ID space.

PVT does not assume 802.1ah support, making due with the IEEE 802.1ad (Q-in-Q) format. However, since the destination MAC address is not under provider control, PVT ignores it, and forwards based on the ingress port and VLAN ID alone, hence the name “VLAN cross-connect”. Since the space of VLAN identifiers is limited by the 12 bits of the VLAN field, VLAN IDs are swapped (similar to swapping in ATM and MPLS), making the VLAN ID only locally significant. If more than 4095 VLAN identifiers are needed per port, there are proposals to form a 24-bit identifier by collecting 12 bits from two VLAN tags, or to use an ISID format identifier in an 802.1ad frame. Once set up, Y.1731 OAM can be used to maintain the connection. PVT can coexist with native Ethernet bridging by partitioning the VLAN ID space.

T-MPLS mandates a departure from Ethernet format in that it prepends an MPLS label, formally making the Ethernet service into a client of an MPLS network. This “layer 1.5” MPLS network is constrained to be pure CO by disallowing PHP, path merging, and PHP. The MPLS stack of locally significant labels is used instead of Ethernet MAC addressing. Once set up, Y.1711 OAM is used to maintain the connection. T-MPLS cannot coexist with standard IP/MPLS (layer 2.5 MPLS), although interworking mechanisms are being considered.

Which one of these three approaches will prevail? They each have their merits and disadvantages, supporters and detractors. We’ll just have to wait and see…