HN Explorer

19 comments

[ 0.98 ms ] story [ 86.2 ms ] thread
ch4s3
Wow, I knew Gin was supposed to be faster than martini, but seeing that large of a spread was pretty interesting.
beckler
Not totally sure, but I think Gin is faster because it uses HttpRouter under the covers.
Yadi
I know this is a dumb question, but is there a conclusion to the benchmarks that which one would be more ideal to use?

I'm just learning GO, that is why I'm asking.

Zikes
I think the tl;dr at the top of the page answers that the best:

    HttpRouter is the king of performance, scoring first on ALL tests. 
    2nd place goes to ace if you don’t need middleware, and Gin if you do. 
    3rd place goes to Goji which is mature, stable and all things considered awesome.
[dead] Yadi
I know this is a dumb question, but is there a conclusion to the benchmarks that which one would be more ideal to use?

I'm just learning GO, that is why I'm asking.

Inflatablewoman
Very interesting. Thanks for posting this. I have been using TigerTonic, but will give Gin a go after reading this.
maximveksler
I hope to not disappoint you, or us as a matter of fact :) hit us up on @hellodebug we will be building this open sourced.
jsherer
Genuinely curious, are there any cases of production Go environments where routing is the bottleneck of their HTTP services? I'd think any kind of data access in the service would be the bottleneck, not routing.
elmin
This is my question as well. With the exception of some very specialized applications, I can't see why any of this would have a significant impact on the real world performance of actual services.
elithrar
Extremely few. 99%+ of your bottleneck will be in template rendering or DB access. Routing is so fast as to not matter for even "large" web applications.
VanillaCafe
The way I look at it is composing benchmarks to estimate the upper bound on the number of pages that can be served per second.

So for GITHUBALL for instance:

* BenchmarkHttpRouter_GithubAll at 66732 ns/op serves at most 15k pages/second/server, versus say

* BenchmarkGoji_GithubAll at 752515 ns/op serves at most 1.3k pages/second/server

Or for GPLUSALL

* BenchmarkHttpRouter_GPlusAll at 3258 ns/op serves at most 310k pages/second/server, versus

* BenchmarkGoji_GPlusAll at 15260 ns/op serves at most 66k pages/second/server

rdtsc
Can imagine it might be it is used as a proxy/gateway to set of servers that then do data access.
falcolas
Silly question, but where's the "plain net/http" benchmark? I haven't found a case where I've missed having an http framework.

Would be interesting to see what the baseline is.

maximveksler
Basically there aren’t much reasons besides security considerations to use net/http multiplexer at all.

The baseline was profiled under STATICALL as the default net/http multiplexer does not support dynamic paths http://golang.org/pkg/net/http/#ServeMux. The benchmark is named BenchmarkHttpServeMux_StaticAll you can read the code at https://github.com/julienschmidt/go-http-routing-benchmark/b.... Frankly, net/http it not a very good http router. The results show that julienschmidt/httprouter trie based implementing outperforms net/http by 60x.