Which is the fastest?

Measuring response times (routing times) for each framework (middleware). Each framework has to have two features; routing and parsing path parameters.

Result

Last update: 2017-11-03

OS: Darwin (version: 17.2.0, arch: x86_64)
CPU Cores: 8

Ranking (Framework)

1. router_cr (crystal)
2. raze (crystal)
3. japronto (python)
4. fasthttprouter (go)
5. nickel (rust)
6. iron (rust)
7. rocket (rust)
8. kemal (crystal)
9. echo (go)
10. gorilla_mux (go)
11. aspnetcore (csharp)
12. iris (go)
13. plug (elixir)
14. phoenix (elixir)
15. vapor (swift)
16. perfect (swift)
17. sanic (python)
18. kitura (swift)
19. akkahttp (scala)
20. clusterexpress (node)
21. express (node)
22. roda (ruby)
23. criollo (objc)
24. sinatra (ruby)
25. rails (ruby)

Ranking (Language)

1. crystal (router_cr)
2. python (japronto)
3. go (fasthttprouter)
4. rust (nickel)
5. csharp (aspnetcore)
6. elixir (plug)
7. swift (vapor)
8. scala (akkahttp)
9. node (clusterexpress)
10. ruby (roda)
11. objc (criollo)

All frameworks

Language (Runtime)	Framework (Middleware)	Max [sec]	Min [sec]	Avg [sec]
ruby	rails	158.707613	157.559875	157.970116
ruby	sinatra	44.278810	44.183649	44.252526
ruby	roda	16.180626	16.132779	16.151732
crystal	kemal	3.233941	3.054319	3.190098
crystal	router_cr	2.432883	2.280620	2.318383
crystal	raze	2.508283	2.407160	2.476729
go	echo	3.798276	3.490771	3.592064
go	gorilla_mux	3.897310	3.561739	3.726176
go	iris	4.016436	3.507674	3.825695
go	fasthttprouter	2.807106	2.716368	2.751400
rust	iron	2.962235	2.717031	2.835935
rust	nickel	2.856162	2.783444	2.834548
rust	rocket	3.223768	3.153677	3.187345
node	express	13.988441	13.727441	13.807823
node	clusterexpress	8.088965	7.502522	7.720822
elixir	plug	4.659650	4.258834	4.477080
elixir	phoenix	4.999001	4.768156	4.844404
swift	vapor	5.538846	5.314528	5.418623
swift	perfect	5.833395	5.570222	5.668958
swift	kitura	6.655281	6.476810	6.548700
scala	akkahttp	8.692796	7.160795	7.619564
csharp	aspnetcore	3.978898	3.616286	3.807356
python	sanic	6.497731	5.388541	5.736287
python	japronto	2.684628	2.553546	2.621803
objc	criollo	34.808480	24.054073	27.575090

Current target frameworks (middlewares)

• Ruby
  • Rails
  • Sinatra
  • Roda
• Crystal
  • Kemal
  • raze
  • router.cr
• Go
  • Echo
  • gorilla-mux
  • iris
  • fasthttprouter
• Rust
  • IRON
  • nickel.rs
  • Rocket (nightly)
• node
  • express
  • express/cluster
• Elixir
  • Plug
  • Phoenix
• Swift
  • Vapor
  • Perfect
  • Kitura
• Scala
  • Akka-Http (Routing DSL)
• C#
  • ASP.NET Core
• Python
  • sanic
  • japronto
• Objective-C
  • Criollo

See Development section when you want to add new languages or frameworks.

The rule

We want to know the response time (routing time), not a usability. So full-stack framework is at a disadvantage.

• Each server has no special logics.
• Each server's executable is named as server_[Lauguage]_[Framework]. (For example, server_ruby_sinatra)
• There are only 3 routes
  • GET '/' return status code 200 with empty body
  • GET '/user/:id' return status code 200 with the id
  • POST '/user' return status code 200 with empty body

Installation

Required environment -> See Current target frameworks(middlewares)

By using Neph

Neph is a modern command line job processor that can be substitute for make command.

To compile servers and benchmarker,

> neph

For each language,

> neph -j ruby

For each framework,

> neph -j rails

See neph.yml

By using make

To compile servers and benchmarker,

> make

For each language,

> make ruby

For each framework,

> make rails

Usage

You can take a benchmark by

> bin/benchmarker

For each language

> bin/benchmarker ruby

For each framework

> bin/benchmarker rails

For comparison (Comparing rails, kemal and router.cr in this example)

> bin/benchmarker rails crystal

If you take it manually, you can run each server by

> bin/server_[Language]_[Framework]

and run client by

> time bin/client

You can set # of threads and # of the loops of the request(there are 3 requests in a loop) by

> time bin/client -t 16 -r 1000

In the above example, 16 threads requests 1000 * 3 times. So 48000 requests are sent in total.

Using Docker

Setup servers by using docker is under WIP. Currently, crystal and ruby servers are supported. For example

docker-compose up rails

Then you can run your client by

time ./bin/client

Development

• Give me PR when you want to add other web frameworks
• Give me PR when you can tuning each framework (under the rule)

Where should I modify when adding new framework

• /[language]/[framework]/[codes] <- Project iteself
• benchmarker/benchmarker.cr <- Adding it as a target to
• README.md <- Adding it as a target framework of the list
• Makefile
• neph.yml (optional)

Anyway, you don't have to care about details since maintainer can fix them after merging it. The result will be updated by maintainer.

Contributing

1. Fork it (https://github.com/tbrand/which_is_the_fastest/fork)
2. Create your feature branch (git checkout -b my-new-feature)
3. Commit your changes (git commit -am 'Add some feature')
4. Push to the branch (git push origin my-new-feature)
5. Create a new Pull Request

Contributors

• tbrand Taichiro Suzuki - creator, maintainer
• OvermindDL1 OvermindDL1 - maintainer

Donate

1AE9P6TUVik1rJGQhaSqGWRk1oAQ3DJnmo