There has been a lot of discussion recently about the importance of supporting east-west traffic in large data center networks. This is supported in a recent article in The Register that highlights marketing data from Cisco* showing not only a large portion of data traffic staying within the data center, but also a large percentage of workloads moving into large cloud data centers. Even today, a single web site click may spawn hundreds of server-to-server transactions that must be completed in time to maintain a good user experience. All of these factors are fueling the drive to large, flat data center networks.
Traditional data center networks have been built using a hierarchy of access, aggregation and core networking gear. In many cases, these networks provide L2 forwarding at the edge, and L3 tunneling in the core. This means that the core must not only support high bandwidth, it must also support complex frame processing, making this a very expensive solution. In addition, these networks exhibit high latency and high latency variation depending on where the east-west traffic is flowing within the network.
The new data centers will employ flat networks consisting of ToR switches at the edge feeding multiple core switches. New products such as the Intel® Ethernet Switch FM6000 family, can enable cost-effective ToR switch designs with advanced L3 tunneling features at the network edge instead on in the core. This frees up the core switch designers to worry only about efficient bandwidth utilization, greatly reducing system cost. When Intel’s Ethernet controllers and switch silicon are used end-to-end in these large data centers, not only is cost reduced, but the high performance and low latency of these products enable efficient east-west data transfers even in the largest cloud data center networks.