This is a collaborative repo maintained by a a number of contributors. Thanks to everyone for making this possible! - Peter Taoussanis
Last updated: TODO
There is no such thing as a one-size-fits-all benchmark. Results will vary dramatically based on your server hardware and the kind of workload you're testing.
So results are organised into named benchmarking profiles that determine wrk config:
| Profile | Connections | Keep alive? |
|---|---|---|
| 1k-keepalive | 32→1024 | ✓ |
| 1k-non-keepalive | 32→1024 | - |
| 60k-keepalive | 10k→60k | ✓ |
| 60k-non-keepalive | 10k→60k | - |
Or see here for older results from before benchmarking profiles were introduced.
Getting started is really easy!
- Clone this repo locally
- Run
start-here.shin the repo root
The particular server hardware, wrk2 config, and server versions will vary between each result set.
Alphabetically:
- Aleph
- http-kit
- Immutant v2
- Jetty Ring adapter
- Jetty 7/8/9, Tomcat 7/8 servlets via lein-servlet
- nginx-clojure compiled into nginx
- Undertow Ring adapter
GitHub pull requests very welcome for:
- Additional web servers
- Updated servers (no snapshot releases please!)
- Server / bench config tuning!
- Please target your PR towards a branch with your GitHub user name.
- Please try to include graph/s when possible. We're looking mostly for relative numbers here so it's not a big deal if the hardware changes between PRs, so long as it's documented and mentioned in any graphs.
- Please try ensure that all servers being benchmarked have a reasonable configuration for your hardware environment. This is especially important if your hardware environment is unusual (e.g. fewer/more cores than usual). If you're unclear on how to adjust some server's configuration for your environment, feel free to open an issue to check that the config you're proposing seems solid to interested maintainers.
If you have any questions, please open an issue on GitHub.
Cheers!