Benchmarking Time Series Databases for IoT at Scale: DolphinDB, InfluxDB, and TimescaleDB
As IoT deployments scale from thousands of devices to millions, the time series database sitting at the core of your data pipeline quietly becomes your biggest liability. Write queues back up. Query latency climbs. What worked at 10,000 devices falls apart at 1,000,000.
To understand where the breaking points actually are, we benchmarked three of the most widely used TSDBs — DolphinDB, InfluxDB, and TimescaleDB — using the TSBS framework, simulating a trucking fleet across five device-scale scenarios ranging from 100 to 10 million devices. We measured write throughput, disk efficiency, and query latency at each level.
DolphinDB came out ahead across the board, and the margin widened with scale. At 10 million devices, it ingested data 17× faster than InfluxDB and 2× faster than TimescaleDB. On query performance, the gap was even larger — tasks that DolphinDB resolved in under a second took the other two databases orders of magnitude longer.
What follows is a breakdown of how we ran the tests, what we found, and what it means if you’re choosing a TSDB for a large-scale IoT deployment.
1. Test Design
All tests used the TSBS standard IoT dataset, modelling a logistics company whose trucks continuously report telemetry. Five scale scenarios were evaluated:
The dataset goes beyond simple time-series streams. Each truck reports two types of records:
- Diagnostics: 1 nanosecond-resolution timestamp, 3 measurements, 8 tag values
- Readings: 1 nanosecond-resolution timestamp, 7 measurements, 8 tag values
- Diagnostics Sample Data
- Readings Sample Data
The test also injects out-of-order data and batch ingestion (simulating trucks coming back online after a period of disconnection) — conditions that are common in real deployments but often glossed over in synthetic benchmarks.
Queries span both real-time monitoring and historical analysis, including vehicle behaviour prediction from time series data. This makes the evaluation representative of the composite workloads IoT platforms actually face.
2. Environment
All three databases ran on the same server to ensure a fair comparison.
Hardware
Software Versions
All three ran in single-node mode.
Key Configuration Parameters
Each database was tuned with parameters appropriate to the workload:
DolphinDB — TSDB block cache enabled and set to 50 GB via setTSDBBlockCacheSize(50).
InfluxDB — Write cache raised to 80 GB to minimise flush-to-disk interruptions; tag value limits removed to accommodate high-cardinality device identifiers; TSI index enabled for high-cardinality performance; full compaction interval set to 30 seconds.
TimescaleDB — Chunk duration tuned per scenario to maintain approximately 12 active chunks:
3. Write Performance
Throughput
DolphinDB delivered the highest ingest rates in every scenario. More importantly, its throughput curve decays far more slowly than InfluxDB and TimescaleDB as device counts climb.
At the 10-million-device scale, DolphinDB sustained 409,000 rows/second. InfluxDB dropped to 24,000 rows/second — a 17× gap. TimescaleDB reached 192,000 rows/second, placing it 2× behind DolphinDB.
Disk Usage
DolphinDB also led on storage efficiency. Its compression ratios outpaced both alternatives across all five scenarios, a direct consequence of the LSM-Tree storage engine architecture it uses under the hood.
Press enter or click to view image in full size
What Drives DolphinDB’s Write Advantage
DolphinDB’s high ingest performance in IoT workloads stems from its deeply optimised LSM-Tree engine, which converts what would otherwise be scattered random writes into efficient sequential writes. This approach fully utilises available disk bandwidth and avoids the write amplification that causes B+ tree–based engines to degrade sharply at scale.
4. Query Performance
Query tests covered 12 categories ranging from last-known-value lookups to multi-truck aggregation and historical trend analysis. We highlight three representative scenarios below.
Scenario 2: 4,000 Devices
At this scale, DolphinDB returned faster results on nearly every query type. Last-value queries (retrieving the most recent reading for a set of devices) consistently resolved in the low milliseconds, thanks to DolphinDB’s dedicated IoTDB engine.
Scenario 2 Results: All Query Types (ms)
A Closer Look: Real-Time vs. Historical Queries
Scenario 3: 100,000 Devices
Scaling to 100,000 devices is where many systems begin to show strain. DolphinDB maintained sub-100 ms latency on high-frequency point queries (last-loc), with performance degradation well below the rate of data growth. The other two databases saw more significant latency increases at this scale.
Scenario 3 Results (ms)
Scenario 5: 10,000,000 Devices
At ten million devices, DolphinDB’s advantage became decisive. Query performance exceeded the alternatives by up to 900%, and last-value queries resolved in approximately 1 second — a level of responsiveness the other databases could not approach at this scale.
Scenario 5 Results (ms)
5. Summary and Recommendations
Across write throughput, query latency, and storage compression, DolphinDB outperformed InfluxDB and TimescaleDB in every scenario tested. The performance gap is not static: it grows as the device fleet grows, which is precisely the direction most IoT deployments are heading.
Selection Guidance
If your device fleet is at or heading toward the 100K–10M range, and your workload combines continuous ingestion, high-concurrency last-value queries, and complex historical analysis, DolphinDB is the recommended choice. Its performance profile at scale directly translates into lower hardware spend, reduced operational complexity, and avoidance of a disruptive re-platform exercise down the road.
If your current fleet is small and growth is uncertain, all three databases are worth evaluating. Factor in operational overhead, ecosystem fit, and your team’s existing tooling before deciding. Running a focused proof-of-concept with DolphinDB as the baseline is a low-risk way to establish a concrete performance and cost comparison.
If you already have complex analytical requirements, prioritise write throughput, compression efficiency, and query performance at scale in your evaluation criteria — the dimensions where this benchmark shows the clearest differentiation.