Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Count-Min sketches for Telemetry: analysis of performance in P4 implementations

José A. Hernández, Davide Scano, Filippo Cugini, Gonzalo Martínez, Natalia Koneva, Alvaro Sánchez-Macián, Óscar González de Dios

TL;DR

This work provides dimensioning rules for CMS at 100~Gb/s and 400~Gb/s and evaluates its performance in a real implementation testbed.

Abstract

Monitoring streams of packets at 100~Gb/s and beyond requires using compact and efficient hashing-techniques like HyperLogLog (HLL) or Count-Min Sketch (CMS). In this work, we evaluate the uses and applications of Count-Min Sketch for Metro Networks employing P4-based packet-optical nodes. We provide dimensioning rules for CMS at 100~Gb/s and 400~Gb/s and evaluate its performance in a real implementation testbed.

Count-Min sketches for Telemetry: analysis of performance in P4 implementations

TL;DR

This work provides dimensioning rules for CMS at 100~Gb/s and 400~Gb/s and evaluates its performance in a real implementation testbed.

Abstract

Monitoring streams of packets at 100~Gb/s and beyond requires using compact and efficient hashing-techniques like HyperLogLog (HLL) or Count-Min Sketch (CMS). In this work, we evaluate the uses and applications of Count-Min Sketch for Metro Networks employing P4-based packet-optical nodes. We provide dimensioning rules for CMS at 100~Gb/s and 400~Gb/s and evaluate its performance in a real implementation testbed.
Paper Structure (8 sections, 2 equations, 4 figures, 1 table)

This paper contains 8 sections, 2 equations, 4 figures, 1 table.

Table of Contents

  1. Introduction
  2. Preliminars
  3. Traffic statistics review
  4. Overview of CMS
  5. Scenario and experiments
  6. Network setting and CMS dimensioning
  7. Experimental validation
  8. Conclusions

Figures (4)

  • Figure 1: Example of a CMS structure, d=3 hash functions and W columns
  • Figure 2: Flow-size distribution: Zipf-like CDF
  • Figure 3: CMS accuracy at estimating top-20 heavy-hitter flows for different configurations of CMS sketch: (top) $d=\{3,5,7\}$ hashes and $w=64$ columns, and (bottom) $d=\{3,5,7\}$ hashes and $w=256$ columns
  • Figure 4: Experimental setup in lab