Data Center Fabric Hype

Why talk about data center fabric? Is this just a buzzword or does it really have value to end users? East-west network traffic is one of the factors driving the discussion.

Why talk about fabric? Is this just a buzzword or does it really have value to end users? A poll at the Gartner Data Center Conference last December showed that the factor driving fabric interest is primarily end users’ interest in virtualization, not a push by vendors. Gartner forecasted that by 2013, 30% of Global 2000 data centers will be fabric enabled.

Data center fabric was also a hot topic at the Ethernet Technology Summit I attended in Santa Clara recently. Based upon my observations, the industry has the following points about fabric:

• East-west traffic is one of the factors driving fabric. East-west network traffic is primarily generated by server to server, or server to storage communications.

• The traditional multi-tier architecture with spanning tree is not suitable to handle east-west traffic.

• A fabric is a sort of mesh interconnect network.

• A fabric can be a large layer 2 network with low and predictable latency.

• 10GbE is for server access and 40/100GbE is for fabric backbone.

However, I haven’t seen the industry agree upon any architecture for fabric, and they probably will never do so. No matter what architecture an end user would choose, a data center fabric needs a high performance, low latency, and long enough reach cabling solution to build its foundation. OM4 multimode fiber is currently the best suitable fiber cabling solution for the fabric in the data center. It has large bandwidth, long reach, is backward compatible for today’s 1GbE and 10GbE networks, as well as providing an upgrade path to 40 and 100GbE.