Re: Benchmarking Web Frameworks, includiing Go, for Games

[bryanturley]

On Tuesday, November 20, 2012 4:27:42 AM UTC-6, in...@helmbold.com wrote:

Maybe you are interested in how well Go performs compared to Frameworks like Netty or Node.js in this benchmark :
http://blog.juicehub.com/2012/11/20/benchmarking-web-frameworks-for-games/

Go performs badly in the first scenario, probably because of not so good JSON and ZIP libraries. But in the second, IO heavy scenario, Go outperforms the competition!

This comparison looks iffy.
Most games would send updates to the game state not the entire game state every single time.
Seems like they should have just called it a web serving benchmark really...

Also they should test vs tip it is faster.

[egon]

They use random json there anyway, so it really doesn't matter for the performance comparison. There's a comment that shows the tip performance. For comparison it's probably better to use the current stable version, dev version can contain bugs that increase the performance. 

 

[Dave Cheney]

I refuse to acknowledge benchmarks taken on Amazon's overloaded virtual infrastructure, using known broken tools like ab. Go may be the best performer, or the worse, but it is not possible to know with this setup.

--
 
 

[Ziad Hatahet]

On Tue, Nov 20, 2012 at 12:14 PM, Dave Cheney <da...@cheney.net> wrote:

I refuse to acknowledge benchmarks taken on Amazon's overloaded virtual infrastructure, using known broken tools like ab. Go may be the best performer, or the worse, but it is not possible to know with this setup.

Out of curiosity, what is so broken about ab?

--
Ziad

[Dave Cheney]

You'll find references on golang-nuts and golang-dev

[minux]

For example, ab (without -k) issues pipelined requests using http version 1.0 without

proper keep-alive headers.

[Dave Cheney]

I use siege currently with the -b flag

I the past I have used httperf and found it very good, but needs more
support scripting around it. All of which I have lost to the sands of
time.