The past three years have been very noisy on the data centre fabric and architecture front. Every quarter seems to bring about a new convergence blueprint from another vendor - and a variety from one or two.
FCoE with Cisco's FabricPath lossless Ethernet software on the company's Nexus switches forms the network backbone. But Cisco is also looking to unify compute and storage with the network through its Unified Computing System (UCS), which includes Cisco blade servers with memory extensions to accommodate a high number of virtual machines and their service profiles.
FCoE is a standard, but FabricPath and UCS are Cisco-specific innovations. Cisco says FabricPath supports the IETF's TRILL method for solving scalability problems in data center networks, but is also a "superset" of it.
Brocade has some FCoE products but is not as bullish as Cisco on the technology. That's because Brocade has the lion's share of the Fibre Channel SAN market - it would lose Fibre Channel revenue to Cisco, which has 65 percent or more of the Ethernet switching market, if the FCoE market takes off.
No TRILL, please
Brocade also claims its BrocadeOne architecture is standards-compliant in that it also supports TRILL and the IEEE's Data Center Bridging standard for lossless Ethernet. Its fabric element is Virtual Cluster Switching (VCS), which was developed to provide a lossless, low latency, deterministic multi-path Ethernet network.
BrocadeOne seeks to transform physical data center assets and resources into virtual services allocated via software commands rather than physical relocation or deployment of systems.
Juniper's QFabric seeks to flatten and simplify data centre networks to accommodate growth of devices and applications, virtualized servers and storage, and cloud-enabled on-demand access to virtualised pools of IT resources. With QFabric, switches and routers that make up a data center network or interconnected data center networks behave as one logical switch or router, meaning applications are no more than a hop away from requester and server.
The QFabric network, in Juniper's view, is a single-tier construct that eliminates two of the three layers - access, aggregation and core - of the data center network.
Juniper eschews TRILL for this purpose in favour of its own technology, which is based on the company's Virtual Chassis scheme. Virtual Chassis combines physical switches into a single logical device.
Indeed, Juniper has been outspoken in its aversion to TRILL, with CTO and Founder Pradeep Sindhu called it "a solution looking for a problem" and "a joke" for data center networks.
Meanwhile, HP and Enterasys have unveiled switching fabrics that stretch across the enterprise, from the data center to the campus and out to the branch office. Avaya has its VENA blueprint, which is based on Shortest Path Bridging; and Alcatel-Lucent has its Application Fluent Network approach.
Is there any right one?
"Who's best depends on the problem you're trying to solve," says Zeus Kerravala, principal analyst at ZK Research. "If you're focused on the data center and storage convergence; if you want to leverage an existing infrastructure; If you want to do something radical and off the wall. They are all coming at it from different points of strength."
But one thing for sure is that there's a need to create a new type of data center network, Kerravala says.
"Everybody would agree that current Spanning Tree solutions don't scale, don't support virtualized traffic. Shortest Path Bridging and TRILL are immature. So any kind of advanced functionality from the vendor is going to have to come in the form of proprietary features. It's likely to be lock-in initially just because of where the standards are. Most customers will be OK with that. It's unlikely that customers will want to mix and match fabric solutions," he says. "Down the road, they'll look to more standardised solutions."
Find your next job with techworld jobs