Elasticsearch Benchmarks

Results

Hint These diagrams are just provided as an overview. Please go to the Kibana dashboard if you want to analyze the results in more detail.

Warning Don't use those numbers to compare Stateful vs. Serverless!

Those benchmarks are meant to detect regressions in Elasticsearch: we make sure they're stable, not fast.
The Serverless instance types that we use have vastly different performance characteristics than our dedicated servers.

Overview

On this page are the results of the Elasticsearch nightly benchmarks based on the docker image built from Elasticsearch serverless repo as of that point in time.

Benchmarks

Benchmarking Methodology

All benchmarks are run by Rally against the Elasticsearch Serverless repo as of that date.

The benchmarks are intentionally not scalability benchmarks but rather show the performance characteristics of Serverless Elasticsearch and are mainly intended to help the development team spot performance regressions.

Benchmarking Environment

Serverless resources

Benchmarks are run in MKI environment with the default node sizing received by serverless end-users.

As of March 2024 default resources are:

4 GiB memory for indexing pod
8 GiB memory for search pod

History of changes

2024-03-19:

Moved Logging and Wikipedia to MKI environment.

2024-03-07:

Moved Autoscaling benchmark to MKI environment, removed Search After Write benchmark.

2024-02-14:

Moved Frequent Docs Update, Geonames, Geopoint, NYC Taxis, and SO Vector benchmarks to MKI environment.