Carnegie Mellon Researchers Pursue Fast Green Boost for Server Clusters

• By Dian Schaffhauser
• 10/27/09

Server performance may get a green speed boost as a result of research coming out of Carnegie Mellon University and Intel Labs Pittsburgh. Researchers have combined low-power, embedded processors typically used in netbooks with flash memory to create a cluster server architecture that is fast but far more energy-efficient for data-intensive applications than the systems now used by major Internet services.

The cluster, still in the experimental phase, is called FAWN, for Fast Array of Wimpy Nodes. The architecture is able to handle 10 to 100 times as many queries for the same amount of energy as a conventional, disk-based cluster. The pilot setup has 21 nodes, each with a low-cost, low-power off-the-shelf processor and a 4 GB compact flash card. At peak utilization, the cluster operates on less energy than a 100-watt light bulb.

The research team, led by David Andersen, Carnegie Mellon assistant professor of computer science, and Michael Kaminsky, senior research scientist at Intel Labs Pittsburgh, have begun work on a next-generation FAWN cluster. This one is being built with nodes that include Intel's Atom processor, which is used in netbooks and other mobile or low-power applications.

Developing energy-efficient server architectures has become a priority for datacenters, where the cost of electricity now equals or surpasses the cost of the computing machines they're populated with.

"FAWN systems can't replace all of the servers in a datacenter, but they work really well for key-value storage systems, which need to access relatively small bits of information quickly," Andersen said. Key-value storage systems maintain data that's used by visitors as they maneuver through a Web site, such as a shopping site that maintains a customer's shopping cart.

Flash memory is faster than hard disks and cheaper than dynamic random access memory (DRAM) chips and consumes less power than either. The research team explained that while low-power processors aren't the fastest available, the FAWN architecture can use them efficiently by balancing their performance with input/output bandwidth.

"FAWN will probably never be a good option for challenging real-time applications such as high-end gaming," Kaminsky said. "But we've shown it is a cost-effective, energy efficient approach to designing key-value storage systems and we are now working to extend the approach to applications such as large-scale data analysis."

The work was funded in part by donations from Network Appliance, Google, and Intel and a grant from the National Science Foundation.