From Earlham Cluster Department
Often the best way to compare two or more computational resources is to analyze how quickly and efficiently they can each solve a given problem. Some of these applications give better light into the resource's performance than others. Be prepared to use some common benchmarking tools.
Top 500 is the canonical list of the fastest computers in the world, with the most recent version of it released on Sunday. How do they figure out how to rank the machines? They measure how much work each machine can do. In order to make sure every center measures their machine in the same way, they're asked to run the same benchmarking program. For the top500 list, this is HPL, High Performance Linpack.
You should compile and run the version of HPL that is shipped with LittleFe/BCCD. The goal is to produce the fastest configuration for the LittleFe/BCCD platform that you can. This run will need to be at least 30 minutes long. Note that you can download and install e.g. ATLAS to significantly improve the performance of HPL.
For your deliverable, you should give us your:
- HPL.out; and
- a write up, potentially addressing these issues:
- What do all the numbers in HPL.dat mean?
- How do the results in HPL.out connect to those numbers?
- How do you think you could have gotten better results, given more time?