Understanding stress conditions of a haproxy setups

Hi,

I'm trying to optimize my haproxy and servers configuration, but I'm having problem when I want to understand what's failing or limiting the configuration.

I'm using autobench with httperf to generate up to 3000 to 5000 request/second; I have a machine with haproxy, and two others with a nginx server each. And I have 3 computers using httperf to stress the application, all synchronized by autobench.

My problem is that autobench reports that it could only achieve up to 2200 requests/second, when I requested 3000 or 4000. So I look for dropped connections, or failed requests in haproxy log and I don't find any.

I already checked my test setup, so I check that each httperf client is able to substain a 2000 request/second, and I have 3 machines with httperf, so reaching that level shouldn't be a problem for them.

I have the tcplog option activated, and the termination_state of the log is always '--' so, no problems there.

I don't know where to look, so, I was wondering if anyone could tip me on what to look, and where to look for the reason that's limiting the request/second.

Thanks in advance!
Mariano

PD: here's the haproxy.conf

global

        maxconn 32000
        ulimit-n 65536
        log localhost user
	daemon


	#nbproc      2 # Number of processing cores. Dual Dual-core Opteron

is 4 cores for example.
	user haproxy
	group haproxy
        stats socket /tmp/haproxy-socket
	chroot /var/chroot/haproxy
	pidfile /var/run/haproxy-private.pid


defaults
	mode	    tcp
        log         global
        option      tcplog
	clitimeout  60000
	srvtimeout  30000
	contimeout  4000


	#option      httpclose # Disable Keepalive


listen	bind 0.0.0.0:80
	balance	roundrobin # Load Balancing algorithm
	server server1 10.251.121.219:81 weight 1 maxconn 2040 check
	server server1 10.250.131.143:81 weight 1 maxconn 2040 check