The data center fabric debate that dominated the industry during the past couple of years has been eclipsed by SDN and OpenFlow. But vendors haven’t stopped moving forward with various fabric alternatives, including TRILL (Transparent Interconnect of Lots of Links), sponsored by the IETF, and SPB (Shortest Path Bridging), also known as IEEE 802.1aq.
TRILL has more commercial momentum largely due to the backing of heavyweights such as Brocade and Cisco. Both vendors base their proprietary fabrics, VCS and FabricPath, respectively, on pre-standards TRILL. They promise a simple, software-only update path to the eventual standard.
Given Cisco's market share and the inherent inertia of data center networks, it's easy to assume that the arguably superior SPB protocol is doomed to also-run status: the Betamax to TRILL's VHS. But SPB isn't without its big name sponsors--primarily companies like Alcatel-Lucent and Huawei that target the service provider market. Add to that list Avaya.
Avaya, which has roots in the old, monopoly-era AT&T networking business, has made SPB the centerpiece of its Virtual Enterprise Network Architecture (VENA) Fabric Connect [pdf] fabric. The technology got its most visible test yet, serving as the backbone for this year's InteropNet at one of the largest networking confabs of the year.
The company claims this to be the largest temporary network constructed to date, with 160-Gbps of aggregated bandwidth. Not only did Avaya technology power the network core and wired edge out to the show floor booths, but it also spanned to 11 different IP video surveillance locations. After watching the IP video demo in Avaya's show floor booth, the network's real time streaming performance was flawless.
InteropNet is built by networking pros for networking pros, and in the assessment of Glenn Evans, its lead engineer and architect, Avaya's fabric implementation is up to the task: "Avaya performed flawlessly as the InteropNet backbone provider. The fact that they were able to get the backbone network up and running very quickly, with minimal staff, is a true testament to the power of the Avaya Fabric Connect technology and the quality of Avaya's engineers."
The network core was built using Avaya's largest Virtual Services Platform (VENA/SPB-capable) device, the VSP 9000, along with its ERS 8800 core switches. However, the edge used the 9000's little brother, the just-released VSP 4000, making its debut performance at Interop. Starting at $10,000, the VSP 4000 is designed to extend VENA fabrics to the campus, metro or WAN edge while also doubling as a hub device for the SMB market.
Avaya claims the VSP 4000 is the first fabric-enabled, multiservice access/edge layer device with MPLS-like functionality at significantly lower cost and implementation complexity. Translated: If you're a small business, you can run multiple L3 VRF (virtual router and forwarding) instances in the same device.
For example, several totally isolated networks can share a single Ethernet pipe, meaning retailers can use the same physical link--say, a metro Ethernet or broadband connection for their guest public Wi-Fi, company private intranet and dedicated PCI DSS-compliant payment network. And because VENA extends L2 SPB to L3 using IS-IS, VRFs can extend across WAN links. VENA's SPB implementation also simplifies network design and deployment because new links or devices are automatically discovered and incorporated into its network management layer.
InteropNet also provided ample opportunity to demonstrate SPB's interoperability as its various elements were provided by over 20 vendors ranging from wireless vendors such as Xirrus to core network suppliers such as Brocade, F5 and HP. Avaya used Interop to do a formal SPB interoperability validation where its VSP 9000 was linked with an Alcatel-Lucent OmniSwitch 6900, HP 12500 Switch Series and tested with a Spirent TestCenter.
Data center fabrics were largely developed to deal with changing network traffic patterns (as internal, east-west traffic, became more important and network engineers sought to avoid hitting the network core with every connection) and more complex topologies (with more servers and storage nodes and an explosion of VMs and virtual switches).
By all accounts, SPB--with support for multiservice L3 VRFs, dynamic topology discovery and low-latency, multipath topologies--succeeds on both counts. This year's InteropNet demonstrated that, ideologically motivated debating points aside, SPB just works.Kurt Marko is an InformationWeek and Network Computing contributor and IT industry veteran, pursuing his passion for communications after a varied career that has spanned virtually the entire high-tech food chain from chips to systems. Upon graduating from Stanford University ... View Full Bio