Power Up

The Raspberry Pi is a great education platform for learning data science. The fact that the Raspberry Pi is relatively slow and low-powered works to your advantage, allowing you to see data manipulation methods unfold at a human time scale. Even for experts, one of the best way to write quality software is to build it on slow computers such as Raspberry Pi. 

Recently I’ve been curious about performing data processing on a Raspberry Pi cluster, which presents some unique challenges. One of the most fundamental problems is finding a proper enclosure.

If you’ve played around with connecting more than one Raspberry Pi together, you’ve no doubt encountered the hassle of tangled cords. To connect six Raspberry Pi boards together, you really need a strategy for keeping things neat. This cluster case is designed to solve that particular problem.

Designing a cluster enclosure is harder than you’d think. Coming at this as a data engineer, I learned that building hardware requires a different level of attention to detail. Even a half millimeter offset in the cutting template, for example, could render one of the acrylic tiers useless.

I also learned that hands-on experience is immensely valuable in designing hardware. The cluster case template I’ve hosted at Github has gone through six revisions. Each iteration revealed one or two shortcomings I couldn’t foresee. It was only after cutting out and assembling the boards that mistakes could be identified.

In the following project I’ve included the template designs under the Solderpad Hardware License. The license is equivalent to Apache, meaning you can make and distribute the designs royalty-free. I would encourage you to contribute to the designs with improvements and extensions.

Update 12/07/15: Want an easy way to configure you Raspberry Pi boards into an Apache Spark or Hadoop cluster? Check out PocketCluster for OSX (GitHub). You can configure a cluster with just 5 clicks (video demo here).