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: 2018-01-18

OS: Darwin (version: 17.3.0, arch: x86_64)
CPU Cores: 8

Ranking (Framework)

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

Ranking (Language)

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

All frameworks

Language (Runtime)	Framework (Middleware)	Max [sec]	Min [sec]	Ave [sec]
ruby	rails	169.990758	169.245815	169.500696
ruby	sinatra	48.785529	48.654910	48.711666
ruby	roda	18.727357	18.608974	18.661612
crystal	kemal	4.178043	4.054511	4.129273
crystal	router_cr	3.455170	3.376436	3.427589
crystal	raze	3.562387	3.444175	3.536099
go	echo	5.256643	4.904162	5.048713
go	gorilla_mux	5.390430	4.980591	5.142020
go	iris	5.244560	4.795147	4.989660
go	fasthttprouter	3.715726	3.595522	3.662270
go	gin	5.682742	5.357033	5.502924
rust	iron	3.485740	3.357875	3.412701
rust	nickel	3.485643	3.416737	3.464607
rust	rocket	4.260527	4.139049	4.211128
node	express	13.895781	13.413687	13.619828
node	clusterexpress	9.048196	7.710249	8.209995
elixir	plug	5.599585	5.259777	5.456209
elixir	phoenix	5.693897	5.393461	5.618488
swift	vapor	6.296311	5.985685	6.120673
swift	perfect	6.570743	6.186247	6.362126
swift	kitura	7.643707	7.318455	7.467837
scala	akkahttp	9.675075	7.849877	8.536331
csharp	aspnetcore	5.405356	4.861280	5.128910
python	sanic	7.291824	6.076758	6.445577
python	japronto	3.587456	3.509572	3.544309
nim	jester	27.304759	27.177061	27.240023
objc	criollo	28.879014	27.740031	28.389459

Current target frameworks (middlewares)

• Ruby
  • Rails
  • Sinatra
  • Roda
• Crystal
  • Kemal
  • raze
  • router.cr
• Go
  • Echo
  • gorilla-mux
  • iris
  • fasthttprouter
  • Gin
• 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
  • flask
• Nim
  • Jester
• 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 ruby

For each framework,

> neph rails

See neph.yaml

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 itself
• benchmarker/benchmarker.cr <- Adding it as a target to
• README.md <- Adding it as a target framework of the list
• Makefile
• neph.yaml (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