It been a while since I published an article on this blog. Originally I intended for it to be used as a platform for me to explain complex physics in a way that non-scientists can understand but I quickly ran out of things to explain because my knowledge is so limited. Instead I’m going to start using this to document my personal projects as well as the research I do at SMU.
So, on to my newest project:
I’ve always been fascinated by high performance computing (HPC) and ever since I started using SMU’s supercomputer, ManeFrame, on a regular basis, I’ve always wanted to build my own cluster for personal computations instead of having to wait in a queue with people that are doing actual science.
The limiting factor for me was cost, a single node on a supercomputer costs upwards of $1000, and its not impressive at all to say that I have a 2 node supercomputer cluster. Instead, I’ll be forgoing raw power in favor of being able to properly parallelize tasks among multiple nodes. My plan is to build a 10 node cluster using a bunch of Raspberry Pi 2 Model B devices.
For background, the RPi2B has a quad core ARMv7 processor clocked at 900MHz and 1GB of RAM. The processor can be overclocked to 1GHz, which isn’t a huge bump, but in devices this small, every bit counts. I have no idea what the performance is going to be like, but I’ve spoken to Rob Kalescky, one of the system administrators of ManeFrame, who built a similar 16 node system using the original Raspberry Pi (single core, 512MB RAM) and got 4GFLOPS of computational power. Thats 4 BILLION floating point operations per second. I think the quad core processors in the Pi2 can have a significant advantage over it, so I’m expecting 6-8GFLOPS once the project is done.
The big question here is money. I’m broke as hell, so how can I get someone else to pay for this??