Zipf’s Law says that the frequency of an item is inversely proportional to its rank in a frequency table.
Mathematically:
-
f(r) = frequency of the item ranked r
-
s = Zipf exponent (a skewness factor, typically between 0.5 and 2)
-
Higher s = more skewed distribution
Zipf-like distributions have been empirically observed in:
Web access logs
Video-on-demand services
CDN (Content Delivery Network) traffic
Edge computing systems
- Cloud file sharing and storage platforms
Useful in Simulation and Modeling:
Researchers and system designers use Zipf to simulate realistic user behaviors when testing caching algorithms, load balancing mechanisms, or data placement strategies.
When Zipf Might Not Apply:
-
In systems where access is uniformly random (e.g., randomized testing, early-stage services), Zipf might not be appropriate.
-
If content popularity changes rapidly, additional models (e.g., dynamic Zipf, time-decay models, or Markov chains) may be more realistic.\