In July 2009, a program known as “Cash for Clunkers” began with the hope of nudging an ill economy back toward health and with a mind to removing crappy cars with poor fuel economy from the roads. You had about a month to bring your clunker to a dealership, sign up to get a new car fitting certain criteria, and get $3500 or $4500 for your clunker. You had to have been able to steam into the dealership in the old heap — no dead cars were qualified.

Lots of people took advantage of the deal.

Guess what? We’re doing a similar deal to move the Duke Compute Cluster forward. We’ve had a lot of clunkers, and they exacted a cost on computational performance (our version of miles-per-gallon). It’s time to move forward, and Cores for Clunkers is a one-time chance to trade up, so to speak. Cores for Clunkers will give researchers who originally bought machines years ago a credit towards the purchase of new machines well in excess of the real value of the devices, just like the relatively high values offered for automotive clunkers in 2009. We’re starting with the oldies, and will move toward the less-old oldies as the program proceeds (just like the “Cash for Clunkers” program).

The recent move to new equipment and new data center surroundings has helped. The oldest Clunker Computers were re-represented in newly minted CPU chips in a new Cisco UCS infrastructure back when we made the big move out of the North Building. The upshot is that if you bought a two CPU machine with 4 GB RAM in 2007, that machine now spins virtually on modern chips. And it spins a lot faster. After the move from North, research computing systems administrators powered down 125 old servers, some that were ten years old — museum pieces, actually, with a single CPU and a mere 2 GB of RAM. They were true clunkers with ten-year-old technology.

People who serve as “points of contact” for the first set of clunker machines were contacted in early April with details about the program.

We’re starting with the 125 old machines. We’ll be in touch with researchers who bought other machines less long ago, as long as we can sustain the Cores for Clunkers program. The result will be a speedier cluster with more current computers.