High response time at high concurrent connections

[Yichun Zhang (agentzh)]

Hello!

On Mon, Sep 30, 2013 at 7:25 AM, laltin wrote:
> But looking at tornado logs I expect around 2000reqs/sec. Assuming that each
> request is handled in 2ms one instance can handle 500reqs/sec and with 4
> instances it sholud be 2000req/sec. But it is stuck at 1200reqs/sec I wonder
> why it is stuck at that point?
> Does increasing the number of instances change the result?
>

It depends on where the bottleneck really is :)

You can use the on-CPU flamegraph and off-CPU flamegraph tools to
check the on-CPU time and off-CPU time usage by sampling your related
processes under load. In particular, using the following tools on
Linux:

    https://github.com/agentzh/nginx-systemtap-toolkit#ngx-sample-bt
    https://github.com/agentzh/nginx-systemtap-toolkit#ngx-sample-bt-off-cpu

These tools are very general and not specific to Nginx processes, BTW.

Best regards,
-agentzh