The new data center network is all about efficient scalability, as can be seen in all the recent press about flat networks from the likes of Cisco* and Juniper*. But efficiency must include key attributes such as latency, bandwidth utilization and cost. Latency has become a key metric in data center networks. This is true not only for financial services and clustering applications, but also for applications such as Internet Portal Data Centers (IPDCs), where a simple client web page access may spawn multiple east-west workflows that must complete quickly to maintain a good user experience. Flat networks are really multi-tier chip networks connected in a fat-tree topology. This can cause network hotspots that can be reduced by bandwidth overprovisioning at the cost of extra network equipment and cabling. Switches that provide efficient bandwidth utilization can minimize the size and cost of the network.
Intel now has a family of 10GbE/40GbE switch silicon that has been optimized for the data center. Its L3/L4 forwarding latency is about a factor of 2 lower than comparable products, and its high quality built-in PHY technology eliminates the need for external PHY chips in top-of-rack switch server connections, further reducing latency, board area and power. In addition, Intel® Ethernet switches provide very efficient load distribution and flow control mechanisms, optimizing bandwidth utilization and further reducing network cost. Finally, the new Intel® FM6000 switch family offers FlexPipe™ technology that allows features similar to a programmable ASIC while maintaining full processing bandwidth of 1Bpps under all conditions. Now network administrators don’t need to rip out and replace switching gear when standards such as VEPA, VN-Tag, TRILL or SPB change; they can simply download a new microcode image to the FM6000, which will support all of these as well as emerging standards.