Time Series Database Comparison
Amazon Timestream vs. CrateDB for Industrial Time Series Data
Announced in November, 2018, the Amazon Timestream time series database service is still not generally available as of January 2019. Other than pricing, there is little detail available about the Timestream service features and API. It is said to include a SQL-like adaptive query engine, data compression, data aging/retention policies, and built-in interpolation and smoothing functions.
Amazon Timestream: Costly for Industrial Time Series?
In a detailed time series data pricing example, Amazon explains that Timestream costs ~$329USD per month to receive data from 10 devices, at 1GB of data per day each, to support a dashboard that queries the data 10 times per day. Unfortunately, industrial time series applications are orders of magnitude more intense.
Consider a typical CrateDB time series workload in an industrial IoT factory setting:
10,000 factory devices inserting 600,000 events per minute (864 million per day), being queried 900,000 times per day to support different real-time dashboards. The inserts alone would cost $150,000 per month on Amazon Timestream.
CrateDB is a better fit for industrial time series workloads such as those found in smart factory, vehicle fleet management, and smart city settings:
Timestream pricing does not scale to support large-scale industrial time series workloads. CrateDB is more economical.
Dynamic schemas, text search, and user defined functions enable CrateDB to support for a wide range of industrial time series analyses.
CrateDB's ANSI SQL interface enables easy integration of time series data with other IT data for richer analysis.
Use CrateDB Cloud as a hosted service, or download and run it on premises at the edge for optimal performance & cost.
CrateDB provides stronger, more affordable support for industrial time series & IoT workloads than Amazon Timestream.
|Organizes data by time interval|
|Columnar indexes & compression|
|Runs anywhere||AWS only|
|ANSI SQL interface||"SQL-like" interface|
|Concurrent query & ingest|
|Fit for Industrial time series||Strong-scalable price/performance, data flexibility & integration.||Weak-too costly to scale. Hard to integrate.|