Web Server: Go, D and Rust

Permalink: HTTP NNTP

Bullet

Posted Wed, 16 Jan 2013 05:01:03 GMT

Reply

Hi, guys!

I've just found a kind of "benchmark" here that might be interesting to someone in forum.
The text is in Russian, but diagram is fortunately in English. Let me translate the summarize for you:

Получилось, что Go слил Node, а Node слила D но все вышли за пределы C10K.

It's such that Go lost to Node, Node lost to D, but all got above C10K.

And he didn't make it to build the only available Rust server yet.

Re: Web Server: Go, D and Rust

Permalink: HTTP NNTP

Dicebot

Posted Wed, 16 Jan 2013 11:37:28 GMT in reply to Bullet

Reply

On Wed, 16 Jan 2013 05:01:03 GMT, Bullet wrote:

Hi, guys!

I've just found a kind of "benchmark" here that might be interesting to someone in forum.
The text is in Russian, but diagram is fortunately in English. Let me translate the summarize for you:

Получилось, что Go слил Node, а Node слила D но все вышли за пределы C10K.
It's such that Go lost to Node, Node lost to D, but all got above C10K.

And he didn't make it to build the only available Rust server yet.

Re: Web Server: Go, D and Rust

Permalink: HTTP NNTP

Dicebot

Posted Wed, 16 Jan 2013 11:38:33 GMT in reply to Dicebot

Reply

Unfortunately, he did not publish any measurement methodics, results are kind of useless without them.
I have benchmarked vibe.d vs cowboy myself and results were drastically different.

Re: Web Server: Go, D and Rust

Permalink: HTTP NNTP

Sönke Ludwig

Posted Wed, 16 Jan 2013 20:45:37 GMT in reply to Bullet

Reply

Always nice to see benchmarks that are not too bad at least (even though I agree that without measurement semantics it's actually not meaningful). But I really need to analyze some day why those strange outliers occur for high connection numbers (if they are what I think they are - the overall performance could be much better for a high number of parallel connections if there wasn't always a small fraction of connections that just time out).