https://bz.apache.org/bugzilla/show_bug.cgi?id=65776
--- Comment #15 from Mark Thomas <ma...@apache.org> --- The large drop in throughput I reported in comment #7 has been bothering me so I wanted to look at this again. Testing this is tricky as you can quickly run out of ephemeral ports. Once that happens, test results become unstable. I think this may have happened when I tested this previously. I definitely saw ephemeral port exhaustion when I tried to re-test this. By reducing the test to a single client thread and temporarily reducing the period sockets spend in TIME_WAIT I was able to get the test results to stabilise. I was then able to test the following changes: - skip the check completely on Windows - add a check for 'duplicates must occur within 1s" to avoid false positives from things like heartbeats on idle systems The results were very promising. The performance difference was (probably) marginally lower but I'd need to repeat the test quite a few times (and brush up on my stats) to provide a definitive answer. My engineering estimate view of the results is that the performance impact is negligible so I'll be apply the change shortly. -- You are receiving this mail because: You are the assignee for the bug. --------------------------------------------------------------------- To unsubscribe, e-mail: dev-unsubscr...@tomcat.apache.org For additional commands, e-mail: dev-h...@tomcat.apache.org
