There are two technologies emerging that will rock the data center design foundation to its very core, Cisco's FabricPath and the IETF standard, Transparent Interconnetion of Lots of Links (TRILL). That's a bold statement, I know, but when we are talking about removing STP from the data center, and still maintaining large L2 domains, it is significant. Data Center engineers need to know the primary differences between them to make critical decisions on how they will design their network and the ramifications of these differences. There is a strong and broad industry effort behind TRILL with multiple companies including Cisco, Intel, Brocade, Stellar Switches and others.
First, let's discuss the commonalities between the two technologies because they are similar and strive to address the same challenges. At a high level, both use the Layer 2 Multi-Pathing (L2MP) nomenclature. Cisco FabricPath and TRILL both perform L2 routing by encapsulating the original frame and placing a new header with new source and destination address information. This new header adds key components like a TTL, to avoid infinite loops like STP permits, Equal Cost Multi-Path (ECMP)routing to maximize traffic distribution and Reverse Path Checks (RPF) for loop avoidance of broadcast and multicast traffic. Additionally both leverage IS-IS as the control plane routing engine. Admit it, how many of you just did a facepalm when you read that last part? Have no fear, it's not as scary as you think.
The differences between the protocols start with the terminology used for the L2MP device. TRILL calls them R-bridges while Cisco FabricPath refers to them as D-bridges. The frame format is different as illustrated below. The Cisco FabricPath frame is on the left while the TRILL frame is on the right. You can see that an additional header is added the TRILL frame referred to as the Next Hop Header and is intended to be used to traverse non L2MP (STP) based networks. Cisco FabricPath addresses this by terminating the L2MP and reverting to traditional STP forwarding.
Comparing the two from a control plane aspect also shows some additional differences in feature functionality. Cisco FabricPath provides additional benefits to the data center designer by incorporating the ability to have multiple topologies. This allows for a more granular control of traffic in the network if desired. Additionally, Cisco FabricPath embeds intelligence in the control plane by using a conversational MAC learning paradigm where the core switches don't learn the end host MAC addresses but rather the switch IDs to simplify the topology, and speed convergence. The final key difference we'll discuss is the VLAN pruning enabled to enable broadcast intelligence. Cisco FabricPath switches will signal to their peers VLANs they have member ports in to dynamically prune unnecessary broadcast traffic from the links.
The burning question is what technology to chose? In classic Cisco fashion, customers will have choices of platforms capable of running either IETF TRILL or Cisco FabricPath. In my opinion, both are compelling technologies but Cisco FabricPath certainly has the edge on feature functionality. Perhaps the IETF will adopt some of the capabilities over time into TRILL but as always, that takes time to develop and mature. Cisco FabricPath is imminent and TRILL deployments are still months away from being implementable.
As with FCoE, I expect customers to test and validate either technology for use in arguably their most critical environment, the data center. Certainly the move from years of STP to L2MP will be significant from both a design and operational aspect and prudence dictates thorough testing. The future designs of data center networks is a changing and I for one think it is for the better. As we say on the interwebs, Giddyup.