harmonic-mean · research

Harmonic mean

The harmonic mean of positive values is their count divided by the sum of their reciprocals.

For n positive values x_i, H = n / sum(1/x_i); the result has the same units as the observations. A weighted harmonic mean is sum(w_i) / sum(w_i/x_i) for positive weights. State the population and weights rather than choosing a mean from the unit alone. [1] [2]

For sequential, nonoverlapping transfers with matching byte accounting, aggregate throughput is total bytes / total elapsed time. Equal byte volumes make this the unweighted harmonic mean of phase rates; unequal volumes require byte weights. Equal durations instead give the arithmetic mean. Concurrent jobs require their common wall-clock interval, not summed overlapping durations. [2] [3]

In a constructed example, 1000 MB at 100 MB/s takes 10 s and 1000 MB at 400 MB/s takes 2.5 s: 2000 MB / 12.5 s = 160 MB/s. Ten seconds at each rate instead gives 5000 MB / 20 s = 250 MB/s. MB means 1000000 bytes; these are accounting assumptions, not measured SSD performance. Include stalls in the elapsed interval when reporting end-to-end throughput. [3]

Relations

related
Geometric mean, Storage throughput

Citations

  1. NIST Dataplot: Harmonic Mean, reciprocal-mean definition (accessed 2026-11-19)
  2. Python statistics: harmonic_mean, equal-distance and weighted rate examples (accessed 2026-11-19)
  3. Hesela: exact synthetic rate and benchmark-ratio examples, executable module (accessed 2026-11-19)

JSON: /data/terms/harmonic-mean.json

Published by Hesela team · Updated

Report a correction · Editorial standards · Reuse guidance

Cite this page
Hesela team. Harmonic mean. Hesela. 2026-11-19. https://hesela.dev/terms/harmonic-mean/