Figure 1: Active And Back-Up Paths - Juniper M10i Application Note

APPLICATION NOTE - Demystifying H-VPLS
In essence you could say that H-VPLS tries to address two different issues:
• The inherent scalability issues of VPLS
• The possibility to extend the VPLS domain to lower end places enabling the use of simpler (cheaper) devices
H-VPLS defines essentially two new roles or functionalities:
• PE-rs: This is basically a PE with all its functionality on the VPLS architecture that runs VPLS with other PE-rs
(following the same VPLS rules) but which also has pseudowires (it can be based on QinQ access, but this will
not be covered) with other devices called MTUs (the access layer essentially).
• MTU-s: This represents the access layer on the H-VPLS architecture and establishes pseudowires to one or two
PE-rs through which VPLS traffic is forwarded.
H-VPLS mode of operation: The mode of operation between PE-rs is like normal VPLS. The operation between MTU-s
and PE-rs is such that the PE-rs treats the pseudowires like access links, or said differently, the split horizon rule does
not apply. If something is received at a PE-rs from an MTU-s, it will be forwarded to the other PE-rs´s and to the other
MTU-s´s connected to the same PE-rs. If something is received at a PE-rs from another PE-rs, it will be forwarded to
the MTU-s´s connected to it through a pseudowire, but not to the other PE-rs (here the split horizon applies).
By means of this mode of operation, H-VPLS tries to make VPLS more scalable, as you will see later in the document.
However, this requires PE-rs's (which would normally be P routers that have no VPLS state) to maintain media access
control (MAC) tables and to perform VPLS operations of learning and flooding. Moreover, a PE-rs has to do this for
all of the MTU-s's that it has. This could lead to data plane scaling problems, especially in terms of the number and
sizes of MAC tables. In summary, H-VPLS creates a control plane hierarchy (in the form of MTU-s and PE-rs), but at
the expense of forcing hierarchy in the data plane as well. Therefore, in the process of solving one scalability problem,
H-VPLS introduces a new scalability problem, and it does not provide solutions for this new problem.
It is important to highlight, however, that the possibility to extend the VPLS domain to cheaper/simpler devices by
establishing pseudowires into a centralized/semi-centralized PE-rs, which is one of the main motivations for using
H-VPLS, is not an exclusive capability of LDP-based H-VPLS. Routers may have this capability regardless of the
signaling protocol that is being used. This is indeed possible with Juniper Networks
2
MTU-s PE-rs
CPE
MTU-s PE-rs
MAC table
LDP active psuedowire
LDP backup psuedowire

Figure 1: Active and back-up paths

MTU-s
PE-rs MTU-s
PE-rs MTU-s
MTU-s
CPE
devices, both using LDP or BGP.
®
Copyright © 2010, Juniper Networks, Inc.