Handling Telematics Data with OpenShift

### Overview Presented at the Open Source Summit Japan 2020 by Ip Sam, a Senior Consultant at Red Hat, this talk addresses the architecture and technical considerations required to process and manage vehicle telematics data using Red Hat OpenShift and edge computing. Telematics merges telecommunications and data processing, gathering real-time telemetry from on-board vehicle sensors (GPS, tire pressure, temperature, door sensors, and engine systems) to optimize operational efficiency, protect drivers, improve customer transparency, and manage regulatory requirements such as fuel tax reporting.

### Key Topics - **Telematics Data Architecture & Cloud Migration:** Transitioning traditional telematics systems into scalable, cloud-native microservices architectures deployed across OpenShift, which provides Kubernetes orchestration, CI/CD automation, and operational readiness tooling. - **Regulatory Complexity (IFTA):** Managing the International Fuel Tax Agreement (IFTA) where dynamic GPS and mileage calculations cross varying state jurisdictions, necessitating low-latency, real-time data inputs. - **Multi-Tier Edge Computing over 5G:** Deploying workloads closer to vehicles via multi-tier architectures ranging from centralized data centers to edge clusters, single edge nodes, and vehicle edge devices. - **Real-Time Edge ML for Road Incident Detection:** Streaming highway condition video/images from truck cameras to edge nodes running artificial neural networks for hazard identification. - **Change Data Capture (CDC) with Debezium:** Using event-driven transaction log replication to eliminate heavy periodic database scans and synchronize telematics data to search indexes and downstream systems.

### Technical Details - **DevOps and Observability:** OpenShift pipelines utilize Jenkins for CI/CD, running unit/integration testing, security and static code analysis, and artifact pushes (e.g., JFrog). System operations leverage Kubernetes liveness/readiness probes, VictorOps, Slack/email alerts, Prometheus metrics, Splunk logging, and Jaeger distributed tracing. - **Edge Computing Topology:** - OpenShift 4.5 edge clusters feature high availability configurations (typically 3 master nodes and 3 worker nodes), stepping down to remote worker setups, single edge node servers, and ultimately constrained vehicle edge devices running single-container workloads over 5G. - When 5G connectivity is intermittent, edge devices implement local caching and retry logic before pushing data upstream. - **Incident Detection Pipeline:** - Truck cameras capture video frames tagged with latitude, longitude, and timestamps. - Data transfers to the edge node with latencies under 200 ms and an upload bandwidth of ~5 MB/s. - Edge nodes run an artificial neural network model (evaluating features such as accident signs, police vehicles, and ambulances) in under 300 ms, triggering alerts if the classification score meets or exceeds a 70% threshold (reaching 97% accuracy). - **Change Data Capture (CDC):** - Implemented via the upstream Debezium project integrated into Spring Boot using the PostgreSQL connector class (`io.debezium`). - As INSERT, UPDATE, or DELETE operations hit the telematics database, Debezium captures transaction logs and relays them through an embedded engine. - The listener maintains file offset backing stores to track processed positions and seamlessly resume after connection interruptions, synchronizing real-time changes directly into Elasticsearch for dashboard updates.

### Takeaways Processing telematics at scale demands an end-to-end distributed topology. Combining OpenShift-managed container workloads across multi-tier edge nodes with 5G connectivity allows compute-intensive operations (such as computer vision hazard analysis) to execute with low latency close to the source. Backed by Change Data Capture (Debezium) and event-driven microservices, organizations can handle massive real-time streams reliably, enabling dynamic routing adjustments, automated regulatory compliance, and robust fault-tolerant operations.