Alert detection in IoT using broadcast state.

on YouTube ↗17m

" summary

Describes how Trackunit uses Apache Flink to detect alerts for over 500K IoT devices governed by 50K+ rules, including single-asset checks, fleet-wide rules, attribute validations, and geo-fencing with advanced scheduling. Explains their design using Flink broadcast state, challenges encountered with broadcast state, scalable geo-fencing implementation, and a runtime mechanism to enable/disable per-rule tracing for debugging distributed stream processing.

AI-written summary · about

" chapters (6)

1. 0:00Start
2. 3:27What is an alert rule
3. 5:29Alert rule predicates
4. 8:39Pros/cons
5. 12:20Debugging
6. 13:20Trace flag

" related talks

• 38m7 reasons to use Apache FlinkJakub Piasecki, Fabian Hueske · Berlin Buzzwords 2019
• 35mEvent Streaming Architecture for Industry 4.0Abdelkrim Hadjidj, Jan Kunigk · Flink Forward Europe 2019
• 31mUnderstanding the Real World: Streaming Analytics for IoTÁlvaro Egea · T3chFest 2016
• 49mDesign patterns for distributed client server systems via IoTEric Shulze, Shyam Krishnamoorthy · AWS re:Invent 2022
• 15mSecurity monitoring for connected devices across OT, IoT, edge, cloudRyan Dsouza · AWS re:Inforce 2023
• 33mFlink for Everyone: Self-Service Data Analytics with StreamPipesPatrick Wiener, Philipp Zehnder · Flink Forward Europe 2019
• 30mThe Hunt for Major League IoT-ICS Threats: A Deep Dive into IoT Threat TerrainMars Cheng, Patrick Kuo · 33rd Annual FIRST Conference (Virtual Event)
• 27mFlowFence: Practical Data Protection for Emerging IoT Application FrameworksDaniel Simionato, Mauro Conti, Earlence Fernandes, Amir Rahmati, Atul Prakash, Justin Paupore · USENIX Security '16