Virtual Private LAN Service

Using AToM/pseudowires as a building block and adding MAC-based forwarding capability to Ethernet over MPLS, a VPLS can be offered. VPLS offers LAN and broadcast domain extensions with multipoint connectivity between sites. Ethernet over MPLS that provides point-to-point Ethernet connectivity is used as a building block. The same signaling procedures can be used to set up the pseudowires between PE devices as in Ethernet over MPLS.

A BGP-based signaling mechanism has also been proposed in the IETF to set up the pseudowires between PEs for VPLs. However, in VPLS on the PE, a virtual switch instance (VSI) is needed. A VSI is a MAC table of the attached devices on the LAN segment or on the remote sites connecting to the VPLS domain. To provide multipoint connectivity, a mesh of pseudowires is set up connecting all PEs that carry the same VSI. The PE forwards the Ethernet frames onto the attached LAN segments that are in the same VPLS domain, or within the same VSI. The PE also forwards frames based on MAC addresses to the pseudowires. This implies that the PE must be capable of learning MAC addresses not just on the local LAN/VLAN segments, but also on the pseudowires. Each PE behaves like a "half bridge" and a PE-pseudowire-PE constitutes a bridge/switch.

All bridging rules apply in the VPLS design and configuration. The PE behaves like an Ethernet switch; in fact, the entire MPLS network behaves like an Ethernet switch, connecting LAN segments at customer sites. This network provides simple forms of connectivity and works well in a metro region with a small number of sites that need to be connected. This might not be the desirable mode of connectivity for a large number of sites in a VPLS domain, though. Scalability of VPLS is discussed in the subsequent section.

Considerations for VPLS

Because the PE devices are forwarding traffic based on the MAC address, as the number of devices grows in the VPLS domain, the MAC table size also grows. The PE must be capable of MAC aging and caching. If the MAC is not found in the MAC table, the frame is broadcast on all ports just as it is done on an Ethernet switch.

With a large number of sites in a VPLS domain and with many devices (PCs, workstations, servers, printers, and so on) in a single VPLS domain, broadcast becomes an issue. Broadcast storms can occur, and these storms can result in network outages. The classic issues of bridged networks are also applicable. Care must be taken to consider the number of sites in the VPLS domain and the number of MAC addresses per VPLS domain as part of the network design.

Loop-free topology can be built using spanning tree or rapid spanning tree in the VPLS domain. However, it is not advisable to run spanning tree over the MPLS network. A full mesh of pseudowires connects PEs with the same VPLS instance, so a split-horizon configuration mode is available. Split-horizon allows PEs not to forward the frames back onto the pseudowire if they learn the MAC address on another pseudowire.

Due to its bridging nature, VPLS has poor scalability. Nevertheless, it is a useful service in small or metro environments. In dynamic provisioning, some level of diverse path routing can be achieved using constraint-based routing mechanisms, although this is not as accurate or comprehensive as the manual provisioning. However, the advantage is that, while provisioning an SPVC, there is no need to touch all the network elements along the path because the diverse routing path is computed by the network elements and set up dynamically via the signaling protocol. This LAN service to IP end points is called IP LAN service (IPLS). Optimizations to VPLS can be done for IPLS for some efficient setup of circuits. However, setup was not the main issue with VPLS; MAC address scale was more of a concern.