"Your choice of industrial time-series database determines your migration costs within a few years." — CrateDB 2026 Industry Report
When building a new data architecture for a manufacturing site, the time-series database (TSDB) is always among the very first decisions. Amid countless online and offline benchmarks, Medium debates, and vendor marketing, TSDB selection generates no small amount of confusion.
As of 2026, the candidates that industrial, manufacturing, and energy sites actually consider number five: InfluxDB (1.x/2.x/3.0), TimescaleDB (Hypercore), QuestDB, TDengine, and the long-established AVEVA PI (formerly Plant Information). These candidates start from completely different design philosophies.
TSDB choice is decided less by "performance" than by "fit with the organization's architecture."
1. The Design Philosophies of the Five Candidates
Each TSDB has a clear design philosophy. Understanding it is the prerequisite for choosing the right combination.
(1) InfluxDB — the pioneer. 1.x used InfluxQL, 2.x used Flux, and 3.0 (GA in April 2025) moved to an Apache Arrow foundation. (2) TimescaleDB — a PostgreSQL extension. Its Hypercore layout keeps recent data in row form (fast inserts) and older data in compressed columnar form (faster analytics). (3) QuestDB — performance-first. Its own SQL implementation and a compact columnar architecture. (4) TDengine — IoT-specialized, with a table-per-device model. (5) AVEVA PI — its design philosophy is "storing and serving process data" itself. Its strength lies in decades of accumulated asset hierarchies, domain features, and calculation engines.
- InfluxDB 3.0 — Apache Arrow, FlightSQL, GA April 2025. Continuous aggregates and materialized views still incomplete
- TimescaleDB — the most natural fit for organizations invested in PostgreSQL
- QuestDB — leads the 2026 benchmarks on raw performance; strong for demanding ingest workloads
- TDengine — IoT/device-optimized architecture, extreme compression
- AVEVA PI — the legacy incumbent; the status-quo choice for organizations that already run it
2. The Truth About Benchmarks — The Trap of Reading the Numbers Raw
The 2026 benchmark published by QuestDB shows dramatic figures. QuestDB claims to be 12-36x faster than InfluxDB 3 Core on ingestion and 43-418x faster on complex analytical queries, and 6-13x faster than TimescaleDB on ingestion and 16-20x faster on complex queries.
These numbers should not be taken at face value, however. Benchmarks are measured on fixed workloads such as TSBS (Time Series Benchmark Suite), which do not match the varied analytics, join patterns, and ad-hoc workloads of real manufacturing environments. Nor do the numbers capture PostgreSQL ecosystem and client-tool compatibility. A recent April 2026 analysis article on dasroot.net likewise recommends "recalculating the scores against your own organizational environment."
Benchmarks only carry meaning inside an organizational context.
3. Recommended Combinations by Organization — Five Scenarios
The concrete patterns in the field look like this:
- Organizations fluent in PostgreSQL/SQL → TimescaleDB Hypercore
- High-ingest, analytics-heavy workloads → QuestDB
- Large IoT device fleets (>100,000) → TDengine
- Process-control-centric organizations (chemicals, power generation, olefins) → keep operating AVEVA PI
- New pilots and broad-scope evaluations → InfluxDB 3.0 or TimescaleDB
4. Case Study — Moving from PI to a Hybrid
A Korean petrochemical company that had operated PI for 20 years kept more than 95% of its collected data in the PI Historian, but as its AI and analytics workloads expanded, it adopted a hybrid strategy of building new applications on a PostgreSQL/TimescaleDB combination.
CrateDB's 2026 industry report names this pattern "PI stays, new workloads go to a TSDB." Migrating the process data of the petrochemical, power, and steel industries off PI is realistically a high-risk project, which is why coexistence is the convenient path. That said, bolting new workloads onto PI should be kept to a minimum.
How PlantPulse Answers
KOPENS PlantPulse is designed not to be locked to a single TSDB. It explicitly supports TimescaleDB, QuestDB, and InfluxDB 3.0 as storage plugins, and organizations running PI can reach their data through PI/AVEVA Connect integration via the same query interface.
The core idea is freedom from data lock-in — giving the organization freedom over migration, hybrid operation, and querying. This also aligns with choosing a TSDB from a data-sovereignty perspective.
Closing
The essence of TSDB selection is not performance but fit with the organization. Is our organization fluent in PostgreSQL? What scale of IoT devices do we run? What are the volume and complexity of our AI and analytics workloads? What assets are already in place? The answers to these questions determine the choice.
Do not decide by chasing benchmark numbers; decide by your own operating conditions. A well-designed combination is stronger than any single "must-have" database. (Related sources: QuestDB "TimescaleDB vs QuestDB 2026 Benchmark," CrateDB "Best Time Series Databases 2026," TDengine TSBS IoT Report, dasroot.net April 2026 analysis, arxiv "Performance Study of TSDB")
© KOPENS — Industrial DataOps & PlantPulse Platform