Network Virtualization using Extreme Fabric Connect

Network Virtualization Using Extreme Fabric Connect
© 2019 Extreme Networks, Inc. All rights reserved. 18
Tip
Benefits of SPB L2 VSNs over EoMPLS and MPLS-VPLS:
• Simple service definition via Service id (I-SID) configured on end-point VLAN
and/or UNI port combination (literally one CLI command), instead of specifying a
Virtual Circuit (VC) on EoMPLS or a Virtual Switch Instance (VSI) on VPLS.
• SPB L2 VSNs natively offer E-LAN (any-to-any), E-LINE (point-point) and E-TREE
(private-VLAN) service types. In contrast, EoMPLS is only point-point and VPLS is
an extension of EoMPLS that dynamically creates a full mesh of EoMPLS circuits to
provide an any-to-any service type. This has a number of disadvantages.
• SPB L2 VSNs have no issue handling packet replication across the Fabric, which is
needed to deliver broadcast and multicast traffic within the service. This is
performed by allocating service-specific shortest-path trees. Whereas VPLS’s
primary shortcoming is that all broadcast and multicast packets need to be
replicated by the ingress PE node multiple times on the same physical interface
(each time with a different MPLS label), which becomes exponentially inefficient as
the number of end-points in the VPLS VSI increases
• SPB combined with Extreme’s SMLT/MLAG/vIST offers an active-active solution
with redundant distribution (BEB-SMLT) nodes. VPLS has a major drawback with
dual homing an access VLAN into 2 redundant distribution PE nodes as this results
in an L2 loop which neither VPLS nor Spanning Tree can prevent. The common
approach is to let only one of the PEs (Primary N-PE) do the traffic forwarding and
the Standby N-PE only forwards traffic in case of failure; Primary and Standby PEs
are usually staggered across the VSI instances
• No need for any BGP (and therefore no need for BGP Route Reflectors either).
• No need for MPLS (and therefore no need for an underlying IP IGP or LDP).
• L2 VSNs can be IP Multicast snoop-enabled with one click per end-point node. This
is simply not possible with VPLS, which will always flood IP multicast traffic with
the above-mentioned flaw that multicast traffic is inefficiently ingress replicated
• Ability to tightly integrate L2 virtualization with L3 virtualization within the same
Fabric architecture. Something that is hard to achieve combining MPLS-VPNs and
MPLS-VPLS.
Table 3 – SPB vs Traditional L2 Virtualization Technologies
Spanning Tree
MPLS VPLS
Fabric Connect
Applicability
Very small networks
Service Provider & Carrier
Networks
Small to Large enterprise
networks, campus, core, IoT
Small service providers
Shortest Path
No
Path always along root tree
Yes
Path can be traffic
engineered
Yes, always
Robustness
Poor
Spanning Tree needs to talk
to block loops, which is then
prone to meltdowns when
things go wrong
Strong
Strong
Service
Scalability
Limited to 4095 VLANs
4 billion if VPLS used with
BGP signalling or auto-
16 million I-SID (24bits)
theoretical limit