Every Time-Series Database Benchmark Ever. 2025

Contact

Home

Pulse Platform

QStudio

Data

kdb+ Tutorials

Blog

Every Time-Series Database Benchmark Ever. 2025<br>Home<br>Data

Updated 2nd January 2025<br>This page will gather ongoing time-series database benchmarks into one location.

Top Benchmarks - Those that are open, comprehensive and thorough

Benchmarks by Date - Sorted Chronologically

The top benchmarks section will focus on time-series databases that are fast, provide time-series joins, time window aggregations and data compression,<br>rather than pull in more general solutions.

Updated 2nd January 2025

Various open-source databases got closer to top performance in clickbench.

See our blog post for the data trends and events of 2024.

It's not a benchmark, but we tested to see if DuckDB can run every query that kdb+ did in a recent KX article.

Updated 10th September 2024

For me, there are three interesting happenings lately:

DuckDB/QuestDB are both now very competitive. (DuckDB in th last year moved 2.7->2.1 in clickbench. QuestDB 24.2->2.7. Lower = better)

ClickBench now includes many parquet/partitioned based timings showing the rise in popularity

The inclusion and rising popularity of embedded databases: chdb and duckdb

Clickbench

Clickbench<br>is a large suite of benchmarks produced by clickhouse themselves. The focus is not time-series but a wide range of queries.<br>They are being very transparent and open, i.e. on some queries they are beaten but the benchmarks only include open source choices.

DatabaseRelative Runtime (Lower = Better)<br>Umbra×1.60<br>ClickHouse ×1.75<br>SelectDB×2.12<br>Doris×2.14<br>DuckDB ×2.19<br>StarRocks×2.36<br>Databend×2.44<br>chDB ×2.55<br>QuestDB ×2.62

Note query 23/28 were removed to get these numbers as the text/regex queries are not of great interest to us.

Top Time-series Benchmarks

To be considered a top benchmark, a combination of factors including repeatability, open-source,<br>reputation, comprehensiveness, thoroughness were considered. This narrowed the top benchmarks to:

benchmarking 4 queries on 1.1 Billion NYC Taxi and Uber Trips - by Mark Litwintschik's. Uses different hardware but thorough writeup of each database features.

Clickbench - 30+ Databases with 42 queries benchmarked. Only includes open-source but highly reproducible.

STAC-M3 - Non-open costly commercial benchmark

H2O.ai - reproducible benchmarking of database-like operations in single-node environment.

Updated Aug 14, 2023

The latest updates were all vendor customized benchmarks for ingestion, each of the vendors benchmarks showed themselves as the fastest at ingestion:<br>questdb, GridDB,<br>TDEngine.

Taxi Ride Benchmarks

Results reproduced from Mark Litwintschik's excellent article.

Mark benchmarked 4 queries against a 1.1 Billion NYC Taxi Trips data set. It had 51 columns was 500 GB in size when in uncompressed CSV format

SetupTotal Query Time<br>(lower = better)<br>kdb+/q & 4 Intel Xeon Phi 7210 CPUs1.0<br>ClickHouse & an Intel Core i9-14900K2.3<br>DuckDB 0.10.0 & an Intel Core i9-14900K2.8<br>Hydrolix & a c5n.9xlarge cluster3.7<br>ClickHouse & a 3 x c5d.9xlarge cluster4.1<br>OmniSci & a 16" MacBook Pro4.3<br>Clickhouse on DoubleCloud, s1-c32-m1285.8<br>BigQuery 8.0<br>Redshift & a 6-node ds2.8xlarge cluster8.0<br>BrytlytDB 1.0 & a 2-node p2.16xlarge cluster13.4<br>ClickHouse & an Intel Core i5 4670K22.2<br>Amazon Athena25.3<br>Elasticsearch (heavily tuned)26.3

STAC-M3

STAC-M3 are highly commercial and results are not published or reproducible.<br>Typically each vendor pays to perform a STAC-M3 benchmark and at the time they use the fastest available hardware humanly possible<br>and a highly optimized configuration. making results extremely difficult to compare. They are however as far as we know, extremely thorough.

Below is a rough overview of the winners doing press releases by year:

System & MachineRelative time (lower is better)<br>2022 - Maykdb+ + Google Cloud improves up to 18x in latest STAC benchmark<br>2022 - FebruaryDDN and Shakti Announce Record Breaking Results on the STAC-M3 Benchmark for Financial Trading Applications<br>2020INFOWeka and KDB 3.6 claim 'Record-Breaking Results on 17 STAC-M3 “Tick Analytics” Benchmarks'<br>2014McObject's eXtremeDB McObject and Lucera Set Records for Market Data Analysis<br>2012McObject's eXtremeDB Financial Edition Sets Records in STAC-M3 Benchmark of Market Data Analysis

h20.ai Reproducible Benchmarks

Aims to benchmark various database-like tools popular in open-source data science.<br>It runs regularly against very latest versions of these packages and automatically updates.<br>They provide this as a service to both developers of these packages and to users.<br>You can find out more about the project in Efficiency in data processing slides and<br>talk made by Matt Dowle on H2OWorld 2019 NYC conference.

All Other Time-series Benchmarks by Date

Rows marked in the below table mean they are produced by a vendor themselves and<br>that you should assume they have chosen a benchmark to highlight their best possible performance.<br>If...