Maxim Konovalov Wrote:
-------------------------------------------------------
> On 11/9/12 7:46 AM, arnoldguo wrote:
> > I have a server running proxy_pass the http request(mainly
> GET,html,jpg
> > etc.) to backend server(20+ server running nginx cache).
> > this server with 32 core CPU,64G RAM,2X 10G NIC,RHEL 6.3,backend
> server with
> > 1000M NIC.
> > When the load raise to 300k connections,3Gbps network traffic,new
> connection
> > can not be serviced,cannot connect to nginx's port,but the system
> load is
> > low(~1.00),and has many free memory,
> > no errorlog in /var/log/message,has some nginx errorlog
> like"upstream timed
> > out to backend server",
> > but the backend's service is normal,load is not hight...
> > i have tune tcp config,can handle 2000K conn,it seems that nginx
> cannot
> > handle so much connection?
> >
> How did you do that and how did you check that the OS can handle 2M
> tcp connections?
>
I use a 3rd nginx PUSH module,and can get 2M nginx connections...

> > nginx config:
> > worker_processes 32;use worker_cpu_affinity;
> > keepalive 32 to backend servers,
> > proxy_buffering on or off has the same result...
> >
> It's almost always about tuning various OS limits not nginx.
>
> As for nginx, this link is a good start:
>
> http://nginx.org/en/docs/ngx_core_module.html#worker_connections
>
> --
> Maxim Konovalov
> +7 (910) 4293178
> http://nginx.com/support.html
>
> _______________________________________________
> nginx mailing list
> nginx@nginx.org
> http://mailman.nginx.org/mailman/listinfo/nginx