Time-Series Ingestion Benchmark

Time-series ingestion benchmarking evaluates how well a time-series database (TSDB) handles the relentless, append-heavy write patterns of monitoring, IoT, and financial data. Systems like InfluxDB, TimescaleDB, Prometheus, QuestDB, and ClickHouse are commonly compared this way. The defining stress is sustained high-rate writes of timestamped points, often across millions of distinct series (high cardinality), combined with time-range queries — a profile very different from OLTP or OLAP benchmarks.

What It Measures

The key metrics are ingest rate (data points or rows written per second), query latency for common time-series patterns (range scans, downsampling, last-value, group-by-time aggregations), compression ratio (on-disk bytes per point, since time-series data compresses well and storage dominates cost), and cardinality handling (how performance degrades as the number of unique series grows). Sustained ingest under concurrent queries is the central concern.

Methodology

The most widely used tool is the Time Series Benchmark Suite (TSBS), originally from the TimescaleDB team. TSBS generates realistic data sets — a DevOps server-monitoring scenario and an IoT fleet scenario — with configurable scale (number of hosts/devices and time span, which together set cardinality). It loads the data while measuring ingest throughput, then runs a catalog of standard queries (single-host metrics, aggregations over many hosts, last-point lookups, complex group-bys) and records latency per query type. You vary cardinality, batch size, and concurrency to map how each engine scales. Storage footprint after load gives the compression ratio. Because TSBS supports many databases through a common generator and query set, it enables fair cross-engine comparison.

How to Interpret Results

Report cardinality, batch size, and write concurrency, since ingest rate is meaningless without them — many systems are fast at low cardinality but degrade sharply as series count rises. Read ingest and query results together: an engine optimized purely for write throughput may answer aggregations slowly. Compression ratio matters for long-retention deployments where storage is the main cost. Always test ingestion and querying concurrently, because production systems do both at once, and isolated numbers overstate capability.

Limitations

Generated data is synthetic and may not match real workload skew, late-arriving points, or out-of-order writes. The benchmark focuses on metrics/IoT patterns and may not represent tracing or log workloads. Results depend heavily on retention, downsampling, and hardware configuration. Use time-series ingestion benchmarks like TSBS to compare write throughput, query latency, and compression under cardinality you expect, then validate with your own data shape. With observability and IoT data volumes growing steadily, a reproducible ingestion benchmark like TSBS is one of the most reliable ways to separate marketing claims from real sustained write-and-query capacity.