ApacheBench doesn't do TLS resumption, so you're forcing a new TLS
handshake with each request. This will kill your performance. ab is a
pretty weak tool ;)

On Thu, Mar 3, 2016 at 7:55 AM, huakaibird <nginx-forum@forum.nginx.org>
wrote:

> Hi,
>
> I want to test the nginx server performance with different server
> configuration (CPU and RAM etc)
>
> I first use apache ab as testing tool, nginx server with 2 CPU and 4G RAM,
> http test could handle 7000 requests/s, cpu usage reach to 30%-40%. But
> https' performace drop dramatically to only 300-400 requests/s and cpu
> usage
> reach to 100%.
>
> Then I also use jemeter and wrk to test, find https' performance is not so
> much less than http, it's only a little than http. For example use wrk test
> http, can reach to 5300 requests/s, while https could reach to 5100
> requests/s and CPU almost same, will not reach to 100%.
>
> So I'm confused, is that because the test tool ab's problem, if it's ab's
> problem, why it's http performance is good?
> If ab's test result is correct, is https' performance is so bad than http?
>
> Posted at Nginx Forum:
> https://forum.nginx.org/read.php?2,265046,265046#msg-265046
>
> _______________________________________________
> nginx mailing list
> nginx@nginx.org
> http://mailman.nginx.org/mailman/listinfo/nginx
>
_______________________________________________
nginx mailing list
nginx@nginx.org
http://mailman.nginx.org/mailman/listinfo/nginx