Cell phone supercomputing II: in the flesh

In a recent blog, I wrote how tiny ARM-based compute nodes were assembled into a 16-way cluster and displayed at the recently concluded SC09 supercomputing conference.

The nodes were put together by Plugcomputer.org and contain a single 1.2GHz Marvell Sheeva Processor (ARM compatible), 512MB RAM, 512MB flash, and one each Gigabit Ethernet and USB sockets. My blog was scintillating and packed full of useful information, but it was not a substitute for seeing this beast in person.

I found their booth as the show was winding down and spent a few minutes talking with Bill Boaz from System Fabric Works, a storage OEM and integrator, who graciously walked me through the system – even pulling the motherboard from a node to show the innards.

To me, this is an interesting demonstration project showing how extremely low power devices (about 5 watts per node) can be pinned together to do useful work. Although there wasn’t a real workload running on the cluster, Bill estimated that each node had roughly the integer performance of an Intel Atom processor – which is more than I expected, to be honest.

I had video rolling throughout our chat, so you can see just how small this cluster (“Pluster,” given the Plug Computer nodes) really is.