The Story of HyperLogLog: How Flajolet Processed Streams with Coin Flips

TL;DR

This paper provides a historical overview of data streaming algorithms, focusing on Philippe Flajolet's contributions, especially the development of HyperLogLog, highlighting the evolution of streaming algorithms through personal anecdotes and archival records.

Contribution

It offers a comprehensive historical perspective on Flajolet's work in data streaming algorithms, emphasizing the development of HyperLogLog and its significance.

Findings

Historical context of HyperLogLog development

Insights into Flajolet's approach to streaming algorithms

Archival records shedding light on algorithm evolution

Abstract

This article is a historical introduction to data streaming algorithms that was written as a companion piece to the talk "How Philippe Flipped Coins to Count Data", given on December 16th, 2011, in the context of the conference in honor of "Philippe Flajolet and Analytic Combinatorics." The narrative was pieced together through conversations with Philippe Flajolet during my Ph.D. thesis under his supervision, as well as several conversations with collaborators after his death. In particular, I am deeply indebted to Nigel Martin for his archival records. This article is intended to serve as an introductory text presenting Flajolet's data streaming articles in a projected set of complete works.