A Better Approach To Scale Vpls Services; Control Plane Perspective; Forwarding Plane Perspective; Figure 5: Point-To-Multipoint Lsp - Juniper M10i Application Note

APPLICATION NOTE - Demystifying H-VPLS

A Better Approach to Scale VPLS Services

Scalability for any type of service or technology can be viewed from different perspectives, and the more that this is
analyzed, the more likely you will achieve the result required. A service provider running VPLS technology or willing
to provide VPLS services should care about scaling from the following perspectives:
• Control plane
• Forwarding plane/bandwidth efficiency
• Service reachability (regional, national, international)

Control Plane Perspective

LDP-based VPLS clearly has limitations in its scalability, due to the fact that it is point to point in nature. In
LDP-based VPLS, the LDP full mesh is tightly integrated with the pseudowire full mesh required on the forwarding
plane. This way, in order to scale the control plane, VPLS requires that the forwarding plane be modified, effectively
breaking the forwarding plane into different domains (which is what is done by means of the H-VPLS PE-rs´s). The
control plane is also separated into different domains, reducing the N-square stress.
In BGP-based VPLS, the control plane and the forwarding plane are decoupled (see section 3.6 on RFC 4761
Hierarchical BGP VPLS). This means that it is possible to implement mechanisms in the control plane itself to make
it scale (use of route reflectors is the most common one) without introducing any change in the forwarding plane.
BGP is inherently hierarchical, so the service provider can take advantage of this to scale the control plane. BGP
VPLS does not require the route reflectors to maintain MAC tables and do VPLS data plane operations.

Forwarding Plane Perspective

Scaling the forwarding plane will typically be forced by the bandwidth consumption that the ingress replication on
VPLS implies. As already described before, we believe the most efficient way to scale the forwarding plane is to add
the capability to VPLS to use point-to-multipoint communication label switched paths (LSPs) as opposed to using
H-VPLS mechanisms.
If you really think about the cause of the problem itself, you see that it is because of the point-to-point nature of the
MPLS LSPs used by VPLS. However, MPLS already resolved this some time ago by the definition and implementation of
point-to-multipoint LSPs, either based on LDP or RSVP-TE (each one with its advantages). So why not go to the origin
of the problem and solve it? This is what Juniper has been proposing by integrating the use of point-to-multipoint LSPs
with VPLS in such a way that broadcast, multicast, and unknown unicast will be forwarded using point-to-multipoint
LSP, which will only be replicated on the network where the network really requires it. By adding this capability to
VPLS, network and bandwidth efficiency will be achieved, while at the same time preventing the increase of the network
complexity with unnecessary MAC/bridging nodes and/or protocols such as Spanning Tree Protocol.
6
Point-to-Multipoint LSP
PE
CPE
CPE
PE
MAC table

Figure 5: Point-to-Multipoint LSP

CPE
PE
P P PE
P P PE
PE
CPE
CPE
CPE
Copyright © 2010, Juniper Networks, Inc.