Researcher at Boise State University Creates Raspberry Pi-Based Supercomputer
What’s a fellow to do when the campus Beowulf cluster isn’t quite ideal for his Electrical and Computer Engineering dissertation project? Build one of his own, of course. Boise State’s Joshua Kiepert did just that, and he used 32 Raspberry Pi boards to do it.
Kiepert’s dissertation is focused on “developing a novel data sharing system for wireless sensor networks to facilitate in-network collaborative processing of sensor data”, and his testing method required a distributed simulation over a LAN, and the Beowulf cluster in Boise State’s “MetaGeek Lab”, aka the Onyx lab, seemed ideal for the task. The Onyx cluster sports 32 nodes, each of which is equipped with an Intel Xeon E3-1225 (3.1GHz) quad-core processor and 8GB of RAM.
However, Kiepert ran into a couple of problems with the Onyx cluster. One was the possibility of downtime, and another was the issue of needing to reply on the admin to allow his to install the software he needed. With his own RPiCluster, he avoided those inconveniences and did so for comparatively little money. He noted that each Onyx node cost between $1,000 and $1,500, whereas each Raspberry Pi board (with an 8GB SD card) costs about $45. Thus, a 32-node RPiCluster was a cost-effective solution.
Of course, there’s really no performance comparison between his RPiCluster and the Onyx cluster, and he had to work with an ARM-based system instead of x86, but for his purposes the Raspberry Pi solution was ideal. It’s a great idea, and it’s one that many others may find useful.
Kiepert gets 100 extra bonus points for outfitting the cluster with some baller lights, because hey, if you’re DIY-ing a supercomputer, why not.
Kiepert’s dissertation is focused on “developing a novel data sharing system for wireless sensor networks to facilitate in-network collaborative processing of sensor data”, and his testing method required a distributed simulation over a LAN, and the Beowulf cluster in Boise State’s “MetaGeek Lab”, aka the Onyx lab, seemed ideal for the task. The Onyx cluster sports 32 nodes, each of which is equipped with an Intel Xeon E3-1225 (3.1GHz) quad-core processor and 8GB of RAM.
However, Kiepert ran into a couple of problems with the Onyx cluster. One was the possibility of downtime, and another was the issue of needing to reply on the admin to allow his to install the software he needed. With his own RPiCluster, he avoided those inconveniences and did so for comparatively little money. He noted that each Onyx node cost between $1,000 and $1,500, whereas each Raspberry Pi board (with an 8GB SD card) costs about $45. Thus, a 32-node RPiCluster was a cost-effective solution.
Of course, there’s really no performance comparison between his RPiCluster and the Onyx cluster, and he had to work with an ARM-based system instead of x86, but for his purposes the Raspberry Pi solution was ideal. It’s a great idea, and it’s one that many others may find useful.
Kiepert gets 100 extra bonus points for outfitting the cluster with some baller lights, because hey, if you’re DIY-ing a supercomputer, why not.
