As spanning tree protocol (STP) shows its age in data center networking and Transparent Interconnection of Lots of Links (TRILL) and Shortest Path Bridging (SPB) crawl toward true standardization, network architects on the bleeding edge are left to choose between betting the server farm on a single, generally proprietary fabric vendor and waiting for the fabric wars to end.
With Extreme Networks, Brocade, and Cisco embracing TRILL, Avaya and Alcatel-Lucent supporting SPB and Juniper backing its own QFabric, it could be a long wait.
Despite updates, STP shows it age
First conceived of in 1985 by the then-giant Digital Equipment Corporation (DEC) and standardized in 1990 by IEEE, STP served as the default for routing meshed local area networks in a way that eliminated dangerous loops, corrected meandering paths and allowed redundancy.
But that flexibility came at a cost.“Even with a set of enhancements in various forms over the course of many years, recovery in a spanning tree environment can take a long time,” said Eric Hanselman, research director with the 451 Group. “It can take up to 30 seconds; the original spec took up to 90 seconds when there was a failure, and that’s a long time for a network to be out.”
STP also blocks redundant paths to prevent loops. This constrains bandwidth, which is increasingly in demand with the increase of east west traffic in highly virtualized data centers. This constrains bandwidth, which is increasingly in demand with the increase of east-west traffic in highly virtualized data centers. Driven by these shortcomings, the Internet Engineering Task Force (IETF) and the Institute of Electrical and Electronics Engineers (IEEE) went back to the drawing board and created two competing successor networking standards —TRILL and SPB—aimed at serving the needs of modern data centers.
While both standards have vocal industry and vendor support, the two approaches are generally incompatible; worse, many implementations of the same protocols are not even interoperable.
Pre-standard implementations or STP alternatives: Vendor lock-in
“There’s no such thing as TRILL interoperability,” said Ethan Banks (CCIE #20655), a data center network engineer with an e-learning software company and a blogger at PacketAttack. “If you pick a fabric solution, you are really marrying yourself to that vendor.”
That lack of interoperability—common with new standards—is not something the major vendors have even tried to hide. Brocade’s Jonathan Hudson recently wrote that his company and Cisco, two of the early TRILL leaders, “had no choice but to implement a ’pre-standard’ TRILL variant.” He stated that Cisco was TRILL compliant in the control plane but not the data plane, while Brocade was compliant in the data plane but not the control plane. Both companies have said that as TRILL works its way toward final standardization, after years of committee wrangling, compatibility is on their roadmaps.
“All of our network switches which support FabricPath also work with TRILL,“ said Rajan Panchanathan, Cisco’s director of project management. As with many vendor-driven innovations of emerging standards, what started as proprietary Cisco extensions have often worked their way into the evolving standards, and the same would be true for many of the extensions Cisco has developed with its FabricPath implementation of TRILL, he said.
“Standards bodies take their time to get their heads around something and get consensus,“ he said. “Cisco is focused on solving our customers’ problems. The standards bodies adopt most of or all of Cisco's innovations.”
The compatibility picture is slightly better when it comes to SPB-based fabrics. For one, the technology was built to be backward compatible with modern datacenter switches, easing the pain for those planning a staged upgrade.
But while SPB’s backers have touted successful interoperability trials in a lab setting, the technology has failed to garner as much market adoption or interest as the decidedly incompatible TRILL.
“Shortest path bridging is a service provider play,“ said Banks. “There is some interoperability with the folks doing it, but it’s not going to get a lot of traction in the enterprise space, I don’t think.“
Banks’ Packet Pushers co-host, Greg Ferro, was a little more blunt in a blog post last year, decrying the SPB standard’s capabilities and deriding its major backers, Avaya and Huawei, as bit players.
“I haven’t had a single listener or person ever ask me about SPB,” Ferro wrote. “While I have done some research, it’s obviously not a topic of interest to anyone in my audience.”
Juniper has opted out of both TRILL and SPB in favor of its proprietary QFabric product and is open about why it thinks those standards are overrated.
“There’s an ideal world and there’s a real world,” said Denise Shiffman, vice president of Juniper’s Platform Systems Division. “In the ideal world, fabric standards should have been done 10 years ago to give vendors time to build to the standards. But in the real world, data centers need to upgrade their infrastructure, and they need to control their costs.”
That off-script innovation leaves network engineers like Banks in a bit of a lurch: Choosing a fabric vendor today means being comfortable with long-term commitment in a field where better offerings might spring up overnight—or solid-looking suitors might disappear.
“My advice is to know what you’re getting into,” Banks said. “Brocade has been rumored to be looking for a buyer for a long time. That sort of thing will factor into people’s buying decisions.
Fabric standards are settling, but should you?
Despite all the turmoil, there are positive signs. TRILL and SPB are widely expected to be finalized as standards this year, and the two protocols’ various backers have promised to work on compatibility, with plug fests already being planned (although not, notably, compatibility between the two protocols).
Nick Lippis, founder of Lippis Enterprises Inc. and organizer of one of those events, is bullish about the future.
“I think every enterprise is dealing with a huge challenge: data deluge,” he said. “Everyone knows what’s happening with mobile devices and what they’ve been driving in terms of data requirements and storage requirements.”
That deluge is creating a sense of urgency around more efficient, redundant and available data centers. Independent confirmation of compatibility and performance will help move fabric adoption, Lippis said.
“The market should be heating up in the second part of this year and into 2013,” he said.
For some companies with urgent data-center needs, 2013 might be too long to wait, Banks said, but that is likely to be the minority: bleeding edge financial firms, service providers, the largest enterprises.
Banks looked at FabricPath as an option for his own company, but he found the cost too prohibitive at this point. “So we said we’re not going to use that. We’re going to use the virtual port solution, which isn’t a fabric as such.”
For now, passing on any fabric option, and keeping your eyes open towards the future, is generally an acceptable answer.
“The real question around fabrics is do they deliver the kind of benefits you need with the penalty of single-vendor support?” said Hanselman. “Most people using fabric today are using it on the scale where that single-vendor requirement isn’t a penalty.”
But even if a business can afford to wait and see what happens, it might be a good idea to start talking to vendors about what they can offer—while re-evaluating existing relationships.
“Everybody is doing a little of something different,” said Banks. “You can just bring in the same sales rep you’ve had for five years, but it’s a really smart time to bring in a vendor you’ve never talked to before and say, ’I’m due for a refresh. This is my problem. What do you have to solve this and why?’”
Dig deeper on Network Virtualization Implementation