For completeness, a) make the array very big (much much bigger than the LLC) b) split the work amongst cores (make it parallel with your favorite methodology)