Amazon Benchmark

Benchmark OpenSearch and ElasticSearch with the OpenSearch Benchmark framework.

This project can automatically benchmark OpenSearch and ElasticSearch and report the results.

You can run benchmarks either manually or with Github Actions.

Benchmarking Overview

This project can run the following OpenSearch Benchmark workloads:

• big5
• noaa
• pmc
• nyc_taxis
• vectorsearch
  • OpenSearch has 3 different "workload subtypes" for the three vectorsearch engines it supports:
    • nmslib
    • lucene
    • faiss
  • ElasticSearch only supports the lucene engine.
• noaa_semantic_search
  • This workload is OpenSearch-only, because ElasticSearch does not have the target benchmark features.

Running a benchmark has three main steps:

• Provision infrastructure with Terraform, including machine(s) for the cluster and a client machine.
• Prepare the cluster by ingesting a workload or restoring a snapshot
• Run the workload
  • This step excludes ingest-related operations in the workload due to the previous step.

Results from a benchmark are stored in a remote OpenSearch data store.

Benchmark Manually

You can run a benchmark manually by deploying infrastructure with Terraform and running scripts on the benchmarking client machine.

To run a benchmark manually, see: infra/README.md

Benchmark with Github Actions

You can also run benchmarks with Github Actions. Actions automate all steps of a manual benchmark. Actions are configured to run nightly benchmarks.

To configure and run Github actions, see README-actions.md

View and Report Results

The results are stored in an OpenSearch data store automatically by OpenSearch Benchmark. There is information about how OpenSearch Benchmark configures the data store here.

You can view the results in the data store, or you can export them using our report generator. Our report generator will download results for a time range and create a summary report in Google Sheets.

To use our report generator, see infra/scripts/report-gen/README.md.