I don’t get it, why apachebench result for such a huge amount of request seems had a bottleneck on the several last 10% of the request.
The test run with ab that bundled with Apache HTTP server.
Operating system : FreeBSD 8.0-RELEASE-p2
Web server : apache-2.2.13 (FreeBSD Ports)
I’m running ab with 100.000 total request and 100 concurrent connection and here is the plot result with gnuplot.