Fabric-cating Tomorrow’s Data Center

Demands of VM-heavy networks require east-west traffic flows rather than north-south hierarchical switching.

brocade fabric

A continuing string of customer wins for Brocade’s Ethernet fabric products is focusing attention around a core data center networking issue: hierarchical switching doesn’t cut it with today’s VM-heavy data center architectures.

The Singapore Advanced Research and Education Network (SingAREN) recently launched its SingAREN-Lightwave Internet Exchange (SLIX), a 100 Gigabit per second community network with a Brocade MLXe-4 router at its core.    

Hosted services provider Rackcorp, based in Australia, selected the Brocade Ethernet fabric to scale its networks for the future and automate delivery of more complex customer solutions.

And New Zealand based AOCloud opted for Brocade's Ethernet fabric technology to create a new network architecture for launching convergent cloud solutions.

Ethernet fabrics are heralded as a way of overcoming the limitations and inefficiencies of current networks. According to an IDC paper, “Legacy network designs optimized for north-south traffic are highly inefficient and difficult to scale without significant disruption or delay. These factors have led to a new network architecture — Ethernet fabric — designed specifically to support the automation, efficiency, and agility requirements created for a virtualized cloud environment.”

Earlier this year, in an interview with the Silicon Valley/San Jose Business Journal, Brocade CEO Lloyd Carney offered up a critique of the networking industry: “The networking layer, because of the way it's rolled out and the fact that it’s dominated by so few companies has not been motivated to change.  Now they’ve been forced to change, because in some ways they are so far behind. If you look at the server layer, the efficiency is there but the bottleneck is the network. On the storage layer, the efficiency is there but the bottleneck is the network.”

What’s driving momentum for Ethernet fabric adoption is the growing use of virtualized assets in the data center. “In a highly virtualized environment you need to be running things on fabric to get the performance you need,” Carney said.

Data centers traditionally used Ethernet for TCP/IP server networks and FibreChannel for storage area networks.

But, wrote Zeus Kerravala, founder and principal analyst of ZK Research “The idea of a totally parallel network running a dedicated, single-purpose protocol like FibreChannel confounds network professionals and is one of the reasons why the networking industry has been beating the drum to put a bullet in the protocol…”

Simply moving to Ethernet won’t work though. The traditional tree topology of Ethernet requires traffic to move up and down the tree, or “north-south,” to get to an adjacent rack. But in the virtualized environment, traffic needs to travel “east-west” between servers in multiple racks, which increases latency and restricts bandwidth.

According to Brocade, the network architecture “must move from management of discrete physical devices and physical ports to flows (virtual server to virtual server or virtual server to virtual storage communication). It must be simpler to set up, operate, and scale. And it must be more flexible, highly resilient, and much more VM-aware.”

As IDC summed it up, “In the final analysis, an adaptable and robust network fabric, combined with an automation strategy that expedites and simplifies network provisioning and configuration, provides a solid foundation for datacenter agility in the cloud era.”


Copyright © 2014 IDG Communications, Inc.