In August 2010, two specifications moving through the IETF promised to improve data centre switching and service provider routing. Vendor products already on the market, or about to ship, raised the question of how much they were needed. The two were Transparent Interconnection of Lots of Links (TRILL) for switching and the Locator/ID Separation Protocol (LISP) for routing.
The alternatives came from the two biggest names in the debate. Cisco was preparing FabricPath for its Nexus 7000 switch, and Juniper Networks was promoting Virtual Chassis. Juniper had decided not to pledge support for either IETF spec.
What the two specifications were meant to fix
TRILL targeted the limits of the Spanning Tree protocol, namely scale and slow topology reconvergence. It was meant to be a Layer 2 protocol with link-state routing, so that switches could calculate shortest paths and support multihop routing in large Ethernet and Fibre Channel over Ethernet data centre networks. The forwarding devices were called Routing Bridges, or RBridges. The design was also meant to work alongside existing Spanning Tree gear.
The TRILL work was credited to Joe Touch of the University of Southern California’s Information Sciences Institute and Radia Perlman, who also created Spanning Tree. Cisco’s Donald Eastlake, a principal engineer and co-chair of the IETF working group, described the result in 2010 as roughly Layer 2.5. He said the final RFC would follow once a companion document on IS-IS code points and data structures was settled, which he expected fairly soon.
LISP, meanwhile, aimed at better addressing and load balancing for enterprises using several ISPs. Cisco had originally authored it, and the company acknowledged in 2010 that other techniques could already achieve some of the same goals.
Why vendors questioned the need
Cisco’s FabricPath was meant to do what TRILL set out to do, with more capabilities on top. Juniper went further. Dhritiman Dasqupta, a senior product marketing manager for data centre technologies at the company, said in 2010 that Juniper had been doing this for two and a half years and could find little that TRILL added.
His argument was that TRILL addressed a single problem, Layer 2 multipathing, in isolation, while data centres faced several issues at once. Juniper’s Virtual Chassis joined up to 10 fixed-configuration switches into one logical switch with hundreds of Gigabit Ethernet ports. The company said this collapsed switching layers and favoured east-west traffic between servers over north-south traffic. Juniper’s data sheets for its EX switches and the upcoming Project Stratus data centre switches were therefore unlikely to list TRILL.
Work on TRILL continued regardless. The University of New Hampshire InterOperability Laboratory hosted a TRILL Plugfest in early August 2010, with Broadcom, JDS Uniphase and Oracle among the announced participants. Analysts at the time expected TRILL products in the first half of 2011.
How it played out
The base TRILL specification was published as RFC 6325 in 2011, and a 2016 update, RFC 7780, said development and deployment had since turned up errata and areas needing clarification.
Adoption did not follow the early expectations. A networklessons.com overview of data centre fabrics says TRILL never gained widespread adoption. It describes FabricPath, launched in 2010, as hampered by vendor lock-in and limited hardware support as Cisco shifted its focus to VXLAN. It groups TRILL, FabricPath and Shortest Path Bridging among Layer 2 fabric technologies that failed to catch on widely.
The ipSpace.net blog took a similar view in 2022. In answer to a reader asking whether FabricPath, TRILL or SPB were still alive, it said “Barely”. It described a market fragmented by partial and proprietary solutions after the Layer 2 fabric rush of 2010. It also noted that Brocade had joined the move to EVPN and VXLAN, and that a reader still saw FabricPath deployments, which customers were migrating to VXLAN/EVPN.
LISP followed a longer path. The IETF published the LISP data plane specification as RFC 9300 in October 2022, authored by Dino Farinacci and others including Cisco’s Darrel Lewis. It obsoletes RFC 6830. The document describes separate namespaces for endpoint identifiers and routing locators, and says LISP offers traffic engineering, multihoming and mobility without changes to host stacks or underlay routers.
Both questions from 2010 therefore got different answers. TRILL became a standard that saw little adoption, with vendors favouring their own fabrics and then VXLAN. LISP reached a standards-track revision more than a decade after Cisco’s engineers had conceded that rival techniques overlapped with it.