Step 1: sort each relation   (if not already sorted):

• Cost = 2.b_R (1 + log_N-1(b_R /N))  +  2.b_S (1 + log_N-1(b_S /N))
              (where N = number of memory buffers)

• if every run of values in S fits completely in buffers,
  merge requires single scan,   Cost = b_R + b_S
• if some runs in of values in S are larger than buffers,
  need to re-scan run for each corresponding value from R