Benchmarking memory accesses from each core

Here we use Sushi Roll (my custom, low-noise kernel) to benchmark memory access times from core to each node on a 96-core, 8-node system.