Uma análise comparativa reprodutível de Apache IoTDB, InfluxDB e TimescaleDB para cargas de trabalho IoT

Abstract

The rapid growth of Internet of Things deployments has generated unprecedented volumes of time-indexed sensor data, exposing the limitations of traditional databases and driving the adoption of specialized Time Series Databases (TSDBs). Despite growing interest, most published TSDB benchmarks lack sufficient scripts and configurations for independent replication. This paper presents a controlled, reproducible benchmark comparing Apache IoTDB 1.3.0, InfluxDB 2.7, and TimescaleDB (PostgreSQL 15) across nine IoT workload types at three data-volume scales spanning a hundredfold range in data size, using a fully scripted infrastructure built on Docker Compose and iot-benchmark. Results show that no single database dominates all workloads. IoTDB leads on write throughput (23.6 M pts/s at large scale) and value-filter scaling (5.2× latency growth for a hundredfold increase in data). TimescaleDB provides sub-millisecond downsampling across all scales and the fastest latest-point lookups, backed by full SQL compatibility. InfluxDB offers a balanced middle ground with lower memory requirements. A key finding is that shared-resource benchmark environments produce misleading results: two conclusions from the initial runs were reversed by isolated reruns, underscoring that resource isolation is a prerequisite for valid comparisons.