Network Virtualization using Extreme Fabric Connect

Network Virtualization Using Extreme Fabric Connect
© 2019 Extreme Networks, Inc. All rights reserved. 13
Figure 1 SPBM’s Mac-in-Mac Encapsulation
Note
As a comparison, MPLS always pushes two or more labels onto an Ethernet (or other L2
technology) packet. But not all MPLS labels are the same. The outer-most label is a packet
forwarding label that is used for label switching the packet across the MPLS backbone. (In
an SPB-based Fabric, this role is undertaken by the BVLAN id + BMAC destination
address.)
The inner MPLS labels are purely used as VPN IDs; that is, once the packet has reached its
destination across the MPLS backbone, the inner label will determine whether the payload
is handed off to one or another VRF/VPLS instance. (In an SPB-based Fabric this function
is taken over by the I-SID.)
The ability of IS-IS to compute shortest path trees at the Ethernet layer is also capable of producing
service-specific shortest-path multicast trees for use with L2 service types as well as for IP Multicast
streams. L2 service types need to transport broadcast, (non-snooped) multicast, unknown unicast packets
efficiently over the SPB backbone for delivery at every end-point in the same service. IP Multicast streams
need to be replicated across the fabric only where IGMP receivers exist. For the sake of comparison, the
ability to natively support multicast trees simply does not exist with IP. With IP, dealing with IP Multicast is
complex because it was implemented as an afterthought requiring additional protocols such as PIM-SM
which can be completely eliminated in an SPB-based Ethernet Fabric.
An SPB Fabric consists of two types of nodes, Backbone Edge Bridge (BEB) and Backbone Core Bridge
(BCB). BEBs are generally deployed at the edge of the fabric. BEBs terminate fabric services (L2, L3,
multicast) and provide interfaces to networked devices or other non-fabric network services. BCBs are
generally deployed in the center of the network. BCBs only perform a transport function along the shortest
path towards the destination BMAC and have no knowledge of the transported service types. Only BEB
nodes add and remove Mac-in-Mac encapsulation to traffic as it enters/leaves the service it belongs to.
This is fundamentally different from traditional enterprise networks but draws parallels with MPLS-based
architectures where service configuration is done solely at the edge of the network without any scaling
impact or need to “touch” the core of the network.
Tip
MPLS networks have a similar concept of Provider Edge (PE) nodes and Provider (P)
nodes.