Cisco FabricPath: Enhancing TRILL: Standards Compatible

Cisco's announcement about FabricPath at CiscoLive cleared up the confusion about what FabricPath actually is. Cisco indicated that Nexus customers would have to choose between Cisco's own layer 2 multi-path protocol or TRILL, the other multi-path bridging standard that Cisco is supporting. A choice like that could cause lock-in to Cisco's proprietary protocol and product line. Cisco was also light on details about how FabricPath ehnahces TRILL. We asked Cisco for clarification on these points.

FabricPath is an implementation of the IETF's TRILL protocol and Cisco has committed to support both multi-path protocols on the Nexus product line simultaneously. It will be able interconnect Cisco Nexus switches and non-Cisco switches into a single multi-path network.

Cisco's multi-path protocol is called Layer 2 Multi-Pathing (L2MP). It is based on the IETF's Transparent Interconnection of Lots of Links (TRILL), but it isn't TRILL exactly. TRILL is not ratified by the IETF, but the work has progressed to the point that Cisco is confident that the standards concerning packet encapsulation and framing is far enough along that it can be built into new physical interfaces. The control plane, which manages the routing paths through the layer two networks, are in discussion and could change before the standard is finalized. Cisco would implement any changes via software updates. A Packet Pusher Podcast  with Omar Sultan, senior solution manager, Cisco Data Center Solutions and Greg Ferro, network architect and consultant, fills in some details on the impact of FabricPath on data center strategy.

FabricPath is a suite of technologies that includes TRILL. FabricPath enhances layer 2 multi-pathing by incorporating non-FabricPath switches in the fabric, uses conversational learning of MAC addresses, and computes different topologies based on local policies. These enhancements are the claimed value-add that Cisco has built into FabricPath beyond TRILL.

FabricPath switches can support switches that don't have FabricPath using virtual Port Channels (vPC). vPC is used to bond two uplinks that are connected to different switches, and both uplinks can be used simultaneously. Port channels and link aggregation can bond multiple links together between two switches without introducing a loop in the network.vPC creates a bonded Ethernet link between one switch and two other switches providing a redundant active/active path. Since FabricPath is only available on the  Nexus 7000, that would include all other Nexsus switches and any other switch that supports vPC.FabricPath also supports multiple topologies based on VLAN, which is used for traffic engineering and separation. For example, if you wanted local policy that separates one set of VLANs from another set, you could, with FabricPath, create multiple layer 2 domains and keep them separate. Finally, FabricPath has, Cisco claims, more efficient MAC learning over TRILL. Called "conversational learning," each FabricPath switch learns the MAC address it needs based on conversations, rather than learning all MAC addresses in the domain. Conversational learning is meant to address the threat that an exploding VM population could create far more MAC addresses than a switch could handle.

Since few data centers are at a point where they can benefit from multi-path networks, coming out with FabricPath this early means Cisco gets a jump in their competition, but they also get time to learn the intricacies of a multi-path network. Representatives from Force10 and Voltaire have said that managing a lossless, multi-path network is not like Ethernet where paths are deterministic. There is the potential for areas of congestion to crop up dynamically making troubleshooting difficult.

Cisco's announcement of FabricPath long before the standards are ratified has led to complaints that Cisco wants to lock-in customers. Typically lock-in means the use of proprietary protocols and functions which limit your choice to purchase equipment and which makes replacement difficult. However, Cisco's public statement to support TRILL in conjunction with their own L2MP protocol simultaneously leaves data center managers open to mix and match products as they need to. Cisco and other switch vendors can enhance the operation of TRILL while still remaining standards compliant.

"The subject of lock-in is a tricky one, especially considering customers try to stick to a single switch vendor in the data center. There is a difference between demanding openness and actually taking advantage of it. Lock-in is one of the favorite sticks that Cisco's competitors try to use against it, but in reality features and reliability will always rank higher in customer minds, especially in data center environments where there are simply fewer devices." Steven J. Schuchart Jr, principal analyst,  Current Analysis, said.