hmm, can you help us know how your testing is working? and the system status such as the threading and CPUs data in "top"
if you really need to handle many many connections, for the OS side: time wait in TCP, fd & port limits, memory for the ATS side: connection limit, accept threads & ET_NET threads, NIC irq & CPU usage they need your tweak and test to find out the best for you use case FYI 在 2013-4-18，下午11:54，[email protected] 写道： > Hello. I recently switched from someone else's apache httpd reverse proxy to > trafficserver. ATS seems very capable but I am still learning. > Single reverse proxy. Single origin server. > Base system CentOS 6.4, 2 cpus, 6 Gb, > trafficserver 3.4, base settings except needed changes to allow https, custom > logs > > ATS worked for our test with smaller loads but now with hundreds of incoming > connections some people were not connecting. netstat shows the ip of a system > not connecting as TIME_WAIT. Watching the origin server specific log, it > seems most requests are being processed through ATS constantly. > > I read ATS can handle thousands easily with one server. Is there some base OS > or ATS setting I need to allow this potential? Not ready for clustering yet. >
