Benchmarking and evaluating time-series databases for appliance-level energy data

Abstract

This thesis benchmarks five Time-series databases (TSDBs)—TimescaleDB, ClickHouse, QuestDB, InfluxDB v1.8, and Apache IoTDB, using a controlled dataset simulating 26 appliances across 100 households at minute-level resolution. Extending the TSM-Bench methodology, ingestion throughput, query latency, storage efficiency, and compression characteristics across five workloads are evaluated. Wide-format schemas are compared against narrow-format schemas to understand schema effects on performance.

Results show dramatic performance disparities. ClickHouse achieves faster ingestion than competitors and completes billion-row queries in seconds versus hours for other TSDBs. QuestDB fails catastrophically beyond 684 million rows due to memory exhaustion. TimescaleDB incurs a performance penalty when enabling time-series optimizations for narrow-format data. Apache IoTDB achieves best-in-class storage compression at the cost of slower ingestion. Wide-format schemas universally outperform narrow formats across all TSDBs except ClickHouse, which demonstrates format-agnostic performance.

These findings provide evidence-based TSDB and schema selection guidance for smart-home energy monitoring deployments requiring billion-row scalability.