Show HN: Marple DB – Querying billions of time series datapoints on Parquet+Pg

2 points

7 months ago

Hi HN! I’m Nero from Marple, and we build time series tooling for Aerospace & Automotive. In 2020, we launched our first product: a sort of Grafana on steroids designed for high performance analysis. After integrating with a lot of different TS databases, we kept running into their design limits when trying to deliver the best performance & UX to our users.

That’s why we are launching Marple DB today (https://www.marpledata.com/marple-db).

Marple DB transforms measurements files (CSV, MAT, HDF5, TDMS, …) into a queryable lakehouse. We designed for extreme ingestion performance. A typical customer example: one MDF file can have ~60k channels, going up to 1kHz, for 1 hour. That is a total of 100B datapoints for a single file.

To make this work, we are using a combination of Parquet files on Apache Iceberg + PostgreSQL. Parquet gives us the scalability we need, and Postgres acts as an extremely fast visualisation cache. We provide Python and MATLAB SDKs to talk to these storages in a unified way. Marple DB is available as a paid product only, but we do offer self-managed hosting.

Me and my co-founder MBaert will join the discussion below.

We are happy to hear your questions!