I've written a server and am running some ab tests from 20 EC2 instances. Above a certain level of concurrency (about 30 for each of 20 instances, so 600 "clients" in total), I start getting failed requests. Presumably this is a limitation of my server's performance. The strange thing is that the failed requests are always, without fail, equally distributed over Receive, Length, and Exceptions (and 0 failed requests for Connect). For example, the output of my most recent test contained the following lines:
Failed requests: 576
(Connect: 0, Receive: 192, Length: 192, Exceptions: 192)
Should I expect to see such output on server-performance-related failed requests? If so, why? All of my searches on this issue only bring up failures due to Length, on account of dynamic webpages returning content of unexpected length.
Thanks,
David