The spanning tree protocol is a long standing standard for networking that has come under fire in recent years. Spanning tree doesn’t meet the demands of cloud, i.e., scale, and it doesn’t deliver the latency characteristics that many companies want from large multi-chassis networking deployments. It’s a challenge that was debated this week, during a panel discussion at the Interop Las Vegas conference.
TRILL vs. spanning tree
One way to avoid the challenges of Spanning Tree, suggested by panelist – Francois Tallet, product manager of the Nexus 7000 at Cisco, is to simply connect the network together with a single switch. But that type of deployment only works with large boxes like, not surprisingly, the Cisco Nexus switch that can scale to up to 768 ports.
“Most people don’t want to just rely on a single box,” Tallet admitted. “In a distributed model with multiple boxes there is more stability for troubleshooting.”
That’s where a solution like Cisco’s FabricPath comes in. FabricPath’s goal is make a network look like a single switch, eliminating the need for spanning tree since traffic is all routed inside of the Fabric.
Tallet noted that FabricPath is a pre-standard implementation for transparent interconnection for lots of links, more commonly known as TRILL, broader industry effort that is currently undergoing its own standardization process. According to Tallet, what FabricPath adds on top of TRILL is more active interaction with Layer 3. He noted that in TRILL today, the draft specification only enables networks to route from a single active gateway. He stressed that Cisco is pushing that innovation back into the standard and at some point both will converge.
Shortest path bridging vs. spanning tree
The other key alternative to spanning tree is shortest path bridging (SPB). Panelist Paul Unbehagen, director, Strategy and Standards at Avaya, is also the co-author of SPB standard. In his view, SPB makes the most sense as it’s an evolution of existing IEEE and IETF standards.
In his view, there are only so many ways that you can design IP and Ethernet to work together. What SPB does is it combines the effectiveness of MPLS with efficiency of Ethernet. He stressed that the SPB standard was jointly developed by both the IEEE and IETF. SPB has all the advanced IP knowledge of MPLS and is backwards compatible with spanning tree.
“There are only two protocols that matter, IP and Ethernet,” Unbehagen said.
Since IEEE is very stringent when it comes to backwards compatibility any new standard can’t orphan anything that has become part of the Ethernet standard in the last 40 years. “That means that everything IEEE produces just works,” Unbehagen said.
The challenge for SPB, just as it is with TRILL, is not all vendors are supporting the same technologies. Avaya, Huawei and Alcatel-Lucent are among the big name vendors supporting SPB, while TRILL is backed by Cisco and Brocade.
During a Q&A at the end of the panel, a member of the audience asked if any of the other participating vendors on the panel (Cisco, Extreme Networks, Juniper) would support SPB.
Shehzad Merchant, vice president of Technology at Extreme Networks, said that TRILL versus SPB should not be a religious argument. That said, he did not specifically state whether or not his company would ever support SPB. Cisco’s Tallet, however, was decidedly more direct.
“We have no plans on supporting SPB,” Tallet said. “We’re doing TRILL.”