Fedora 19 performs poorly

[Konstantin Svist]

Note: same message also posted in -ru list

I'm trying to migrate from Fedora 14 (2.6.35.14-97.fc14.x86_64) to
Fedora 19 (3.10.11-200.fc19.x86_64)

worker_processes  40;
events {
    worker_connections  8000;
    use  epoll;
}
http {
    proxy_headers_hash_max_size     8096;   # default was: 512
    proxy_headers_hash_bucket_size  128;    # default was: 64
    variables_hash_max_size         1024;
    variables_hash_bucket_size      128;
    default_type        application/octet-stream;
    sendfile            on;
    keepalive_timeout   65;
    charset             utf-8;
    resolver            127.0.0.1;      # necessary for dynamic upstream resolution
    limit_req_log_level warn;
    proxy_intercept_errors      on;
    server  {
        listen  80;
        location  =  /service_check_nginx  {  echo "nginx";  }
    }
}



Symptoms:

* ab -n1000000 -c1000 'http://localhost/service_check_nginx' (parallel
4x, i.e. 4000 simultaneous connections)
says that some requests take >3sec and/or fails to complete

* netstat -s:
...
1269313 times the listen queue of a socket overflowed
1282868 SYNs to LISTEN sockets dropped
...

grows with speed of 2000/sec, sometimes more


* CPU load is not spread out evenly between workers:

  PID USER      PR  NI    VIRT    RES    SHR S  %CPU %MEM     TIME+ COMMAND
27671 nobody    20   0  418.6m 149.2m   1.1m S  51.1  0.1   0:00.77 nginx: worker process
27685 nobody    20   0  418.6m 149.2m   1.1m S  39.7  0.1   0:01.76 nginx: worker process
27661 nobody    20   0  418.6m 149.2m   1.1m S  22.7  0.1   0:01.63 nginx: worker process
27688 nobody    20   0  418.6m 149.2m   1.2m S  22.7  0.1   0:01.90 nginx: worker process
27697 nobody    20   0  418.6m 149.2m   1.1m S  17.0  0.1   0:00.95 nginx: worker process
27666 nobody    20   0  422.0m 152.3m   1.1m R   7.6  0.1   0:01.50 nginx: worker process
27701 nobody    20   0  419.3m 149.7m   1.1m S   1.9  0.1   0:00.01 nginx: worker process
27650 nobody    20   0  418.6m 149.9m   1.8m S   0.0  0.1   0:03.52 nginx: worker process
27658 nobody    20   0  418.6m 149.2m   1.1m S   0.0  0.1   0:01.30 nginx: worker process
27664 nobody    20   0  419.0m 149.5m   1.1m S   0.0  0.1   0:01.86 nginx: worker process
27669 nobody    20   0  418.6m 149.2m   1.1m S   0.0  0.1   0:00.35 nginx: worker process
27672 nobody    20   0  418.6m 149.2m   1.1m S   0.0  0.1   0:00.23 nginx: worker process

meanwhile, on F14:

  PID USER      PR  NI  VIRT  RES  SHR S %CPU %MEM    TIME+  COMMAND
30042 nobody    20   0 1224m 955m  17m R 41.2  0.4 523:24.45 nginx: worker process
30038 nobody    20   0 1224m 955m  17m S 39.4  0.4 522:24.30 nginx: worker process
30047 nobody    20   0 1224m 955m  17m R 39.4  0.4 520:35.36 nginx: worker process
30053 nobody    20   0 1224m 955m  17m R 39.4  0.4 520:42.77 nginx: worker process
30027 nobody    20   0 1224m 955m  17m S 37.6  0.4 520:55.20 nginx: worker process
30036 nobody    20   0 1224m 955m  18m R 37.6  0.4 525:26.07 nginx: worker process
30037 nobody    20   0 1224m 955m  17m S 37.6  0.4 523:59.09 nginx: worker process
30041 nobody    20   0 1224m 955m  17m R 37.6  0.4 529:31.88 nginx: worker process
30049 nobody    20   0 1224m 954m  17m R 37.6  0.4 519:58.73 nginx: worker process



By the way, if I set worker_connections 800 (worker_processes 40) and
start up "ab -c1000 ..." -- then ab fails with an error  (on F19 only;
fine on F14).

What should I do next?