please suggest performance tweak and the right siege options for load test

[Lucas Rolff]

Have you checked the socket level, and checking kernel log on all 3 
servers (nginx and load balancer) meanwhile doing the test?
It could be that for some reason you reach a limit really fast (We had 
an issue that we reached the nf_conntrack limit at 600 concurrent users 
because we had like 170 requests per page load)


halozen wrote:
> 2 nginx 1.4.6 web servers - ocfs cluster, web root inside mounted LUN
> from SAN storage
> 2 MariaDB 5.5 servers - galera cluster, different network segment than
> nginx web servers
>
> nginx servers each two sockets quad core xeon, 128 gb ram
> Load balanced via F5 load balancer (round-robin, http performance)
>
> Based on my setup above, what options that I should use with siege to
> perform load term to at least 5000 concurrent users?
>
> There is a time when thousands of student storms university's web
> application.
>
> Below is result for 300 concurrent users.
>
> # siege -c 300 -q -t 1m domain.com
>
> siege aborted due to excessive socket failure; you
> can change the failure threshold in $HOME/.siegerc
>
> Transactions:                 370 hits
> Availability:               25.38 %
> Elapsed time:               47.06 secs
> Data transferred:            4.84 MB
> Response time:               20.09 secs
> Transaction rate:            7.86 trans/sec
> Throughput:                0.10 MB/sec
> Concurrency:              157.98
> Successful transactions:         370
> Failed transactions:            1088
> Longest transaction:           30.06
> Shortest transaction:            0.00
>
> Posted at Nginx Forum: http://forum.nginx.org/read.php?2,257373,257373#msg-257373
>
> _______________________________________________
> nginx mailing list
> nginx at nginx.org
> http://mailman.nginx.org/mailman/listinfo/nginx