Read the following two papers:
Truth in SPEC Benchmarks, Computer Architecture News, Vol 23, No. 5, Dec. 1995.
by Mirghafori, Jacoby, and Patterson.
SPEC as a Performance Evaluation Measure, Computer, Vol 28, No. 8, August 1995.
by Giladi and Ahituv
and summarize the arguments for using the harmonic mean instead of the
geometric mean. Pick 5 systems with CPUs from different manufacturers
that have similar SPECfp95 values and calculate new SPEC values using the
harmonic mean. Do this for both SPECfp95 and SPECfp_base95. Do the rankings
of the machines change?