Probabilistic Multiplicity Counting
Switch branches/tags
Nothing to show
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Failed to load latest commit information.
benchmark Allow usage of 64 virtual matrix width Sep 29, 2015
.gitignore Initial commit Aug 27, 2015
LICENSE Update documentation Aug 29, 2015 Updated README Aug 29, 2015
doc.go Add doc.go Aug 29, 2015
pmc_test.go Allow usage of 64 virtual matrix width Sep 29, 2015

Probabilistic Multiplicity Counting Sketch (PMC)


PMC to Count-Min is as HyperLogLog to Bloomfilter

Package pmc provides a Probabilistic Multiplicity Counting Sketch, a novel data structure that is capable of accounting traffic per flow probabilistically, that can be used as an alternative to Count-min sketch. The stream processing algorithm — Probabilistic Multiplicity Counting (PMC) — uses probabilistic counting techniques to determine the approximate multiplicity of each element in large streams. It is particularly well suited for traffic measurements on high-speed communication links and likewise applicable for many other purposes.

Count-Min Sketches hold counters in a matrix-like organization. A big caveat for both Spectral Bloom Filters and Count-Min Sketches is that the maximum multiplicity has to be known a priori quite accurately, to provide large enough counters without wasting too much memory. PMC does not need to know the maximum frequency beforehand, and its counting operation is much simpler.

For details about the algorithm and citations please use this article:

["High-Speed Per-Flow Traffic Measurement with Probabilistic Multiplicity Counting" by Peter Lieven & Björn Scheuermann] (

Example Usage

import ""

sketch, err := pmc.NewSketchForMaxFlows(1000000)

//increment a flow 'flow1' 1000000 times
for i:=0; i<1000000; i++ {

count := sketch.GetEstimate([]byte("flow1"))
// count ==> 994623 (its an approximation)