Ran across a short article this week discussing the experimentation going on right now using Intel’s low-power (and low end) Atom chip inside large cluster and enterprise server deployments.
Some data center channel partners see increased interest in Atom, and major technology vendors including Google and Microsoft are looking at running servers with low-power chips such as Atom. But others remain skeptical, saying that Intel’s Xeon processor will continue to dominate in the data center.The topic of Atom servers is expected to arise at an upcoming Intel channel advisory board meeting, sources said.
There are already experiments in our own community, and the article points out one example (we also know that Dan Reed’s group at MSR has been looking at this, and SGI showed a prototype Atom-based cluster during SC08)
There is definitely experimentation. In October, researchers from Carnegie Mellon University and Intel Labs published a paper called “FAWN: A Fast Array of Wimpy Nodes.” The paper proposed a new “cluster architecture for data-intensive computing.”
“FAWN couples low-power embedded CPUs to small amounts of local flash storage, and balances computation and I/O capabilities to enable efficient, massively parallel access to data,” the paper’s abstract reads. The project employed years-old AMD Geode LX chips, processors primarily used in embedded applications. A second iteration plans to use Atom chips. The project was funded by NetApp, Google and Intel.
Obviously a lot of downside potential here, but in my experience much of the resistance to an idea like this is of the form “that will obviously never work.” This is not a useful statement until it is based on fact, which makes these experiments extremely valuable.