How One NVIDIAN Built a Tiny Server Cluster Out of a Slice of Raspberry Pi
A casual stroll through NVIDIA’s headquarters might not provide you with scents of fresh baked goods, but a multicolored Raspberry Pi server cluster flashing its lights just around the corner might be enough to stop you in your tracks.
You can’t eat it. You might not even see it if you walk by high performance computing (HPC) systems engineer Adam DeConinck’s cubicle too fast, because this cluster stands at a petite five inches. Don’t be fooled, however — this device is the real deal.
Raspberry Pi is a commercially available single-board computer created as a UK charity project to give schools cheap hardware students could tear apart and put back together again without worrying about ruining expensive gear. Since then, the hardy computer has been adapted for plenty of creative uses.
For geeks like Adam, the Raspberry Pi’s ARM CPUs, the same ones used in the vast majority of the world’s mobile devices, provide an alternative to the x86 architecture used by most PCs and servers. Adam – whose day job is building compute clusters on the Amazon Cloud and maintaining NVIDIA’s internal HPC clusters – likes to spend his spare time experimenting with his own, modest-sized cluster.
Although he admits that most people don’t build their own server clusters, which are sets of connected computers that work together as a single, more powerful computer, Adam sees his teensy cluster as a way to play with new ideas. The dinky machine is built from five separate Raspberry Pi ‘servers,’ each with its own tiny ARM CPU. The whole thing uses just 16 watts of power when running at full capacity.
“It’s basically a fun project,” he says. “Whenever I see some cool and interesting computer-related development, the first things I ask myself is, ‘how could I build a cluster out of this?’ I built this cluster because I thought it would be cool to have one on my desk to play around with.”
Adam isn’t the only one who had fun with this project. When Adam asked his wife, Leigh, who’s a non-techie, if she wanted to help create a Lego server rack for the 7 by 9-inch cluster, she quickly agreed. “She thought it was a cool idea,” Adam says.
After ensuring his tiny cluster is optimized for speed and performance, Adam plans to convert it into a Hadoop cluster for Big (or not so big) Data processing, and use it to experiment with other system administration problems.
“Clustering with HPC servers is about solving problems with more than just one computer,” he says. “I build and work with clusters for a living, and one thing I always look for and get excited about is figuring out how you can work with new and interesting hardware in a scientific computing environment.”
Fellow employees agree that the cluster is innovative, but were disappointed when they discovered that there wasn’t yet any NVIDIA hardware in it.
But the future looks plenty green for this little cluster that could. Adam says he wants to get Tegra into it, turning it into a portable demo system to be shown off at conferences. Now that’s something you don’t just blow a raspberry at.