Real-time DDoS anomaly detection

Six years on DDoS Secure at Allot — network behavior anomaly detection and mitigation on live traffic.

At Allot (2014–2020) — which I joined via a startup tiger team that Allot acquired — I spent six years as a software and algorithm engineer on DDoS Secure, a real-time network behavior anomaly detection (NBAD) and mitigation system.

Highlights

  • Full re-architecture — designed and rebuilt the product backend from Ruby to Java, including the PostgreSQL layer and parts of the GUI, while the product stayed live.
  • BGP-based mitigation — built a REST API for the BIRD routing daemon so the system could divert malicious traffic during an attack using BGP.
  • Adaptive detection algorithms — extended pattern-extraction algorithms to track evolving attacks, and applied multiple linear regression to predict a signal’s clean baseline while under attack, so long-duration floods couldn’t poison the detector’s notion of “normal”.
  • Won first prize in the Allot hackathon along the way.

This was my algorithmic apprenticeship: statistics and detection theory applied to hostile, high-volume, real-time data — long before “anomaly detection” had an ML-platform aisle.