This is a simple memory benchmark; see, e.g., "Empirical Evaluation of the CRAY-T3D: A Compiler Perspective" from ISCA 1995. The membench C code basically times

for array A of length L from 4KB to 8MB by 2x
  for stride s  from 4 bytes to L/2 by 2x
    time the following loop
    for i = 0 to L by s
      load A[i]

The raw times can be plotted as a set of lines or as a heat map (I find the latter easier to process visually).

If you are running membench on the totient cluster, you will want to obtain the timings by running qsub membench.pbs rather than by running the membench executable on the head node. The head node and the compute nodes have rather different performance characteristics.