Network Virtualization using Extreme Fabric Connect

Table Of Contents
Network Virtualization Using Extreme Fabric Connect
© 2019 Extreme Networks, Inc. All rights reserved. 21
Note
Extreme’s SMLT clustering supports dual homing of server/hypervisors when the NIC
teaming configuration requires link aggregation on the network side. Multiple homing ( >
2) is only possible when the NIC teaming configuration does not require link aggregation
on the network side.
For IP overlay data center solutions such as VMware’s NSX, SPB can also provide a scalable and robust
underlay infrastructure providing easy to deploy IP multicast for optimized NSX VXLAN deployments,
avoiding any ingress replication requirements on the NSX overlay. This combined with the SPB-VXLAN
Gateway and OVSDB functionality can be leveraged to connect bare metal servers operating in the SPB
fabric underlay into the NSX overlay infrastructure.
Fabric Connect Positioning
Many fabric architectures are offered by different vendors, mostly for use in the data center but in some
cases in the campus as well. This section is intended as an overview of that fabric landscape and to allow
the reader to better understand how the Extreme Fabric Connect architecture compares to other fabric
constructs.
TRILL and Derivatives
TRILL (Transparent Interconnection of Lots of Links) is an Ethernet Fabric standard developed by the IETF
at roughly the same time as SPB was standardized by the IEEE. It is similar to SPBM in many respects as it
also leverages IS-IS directly over Ethernet. It is notably different from SPBM in some design choices that
were made and constitute a trade-off.
TRILL was designed to offer unlimited equal cost multi-pathing where individual TRILL switches can decide
to hash egress traffic across any number of egress interfaces that all provide a lowest equal cost path to the
destination, much like is possible with OSPF. However, this design choice precludes TRILL-based networks
from leveraging the 802.1ag CFM since it becomes impossible for an ingress node to predict what path is
used across the network.
Tip
With SPBM, equal cost shortest paths are only possible across an equal number of
BVLANs. But within each BVLAN there is a single predictable shortest path from source
to destination and from the ingress node 802.1ag CFM can be leveraged to validate that
path.
The same design choice also implies that TRILL-based networks cannot make any assumptions on the
validity of packets arriving on ingress interfaces (since those packets could have been hashed over any
number of equal cost paths). So transient loops can only be dealt with by including a TTL field in the
packets so that looping packets extinguish themselves after looping a few times, much like IP does with
OSPF.
Tip
Within an SPBM BVLAN there can only be one predictable shortest path to any
destination and source alike. SPBM therefore deals with transient loops by applying
Reverse Path Forwarding Checks (RPFC) on every ingress packet. This ensures that
transient loops are instantly suppressed.
The fact that TRILL requires a TTL field for transient loop suppression is one of the reasons why TRILL
defines a completely different packet encapsulation from IEEE 802.1 Mac-in-Mac. The TRILL designers
however made a mistake to simply use the VLAN-ID 12-bit field in the original implementation of TRILL