Hello Nicolas, NC> If one socket is owned by only one thread, it shouldn't cause any NC> problem. Crashes can occur if you are performing some operations on the NC> same socket in different threads.
NC> The reason of socket_recv is not clear to me, it looks like the NC> connection was stopped until all data has been sent, or that more data NC> is being readed than available. I see.. It is actually strange, what I am experiencing is when I am launching stress tester in multi-threaded mode, average response time is about 8-10 seconds and many threads fail. When I do the same on my server (to eliminate network latency), the same number of threads fail but after 50-100 milliseconds waiting. I initially thought that my server doesn't handle the load, but this statistics combined with request processing time about 50 ms (from apache access logs) makes me think that there is something else here. So to summarize: client sockets fail for some strange reason, and it does not seem to be network latency/connection problems. Maybe the problem is on the server actually - but any ideas? -- Best regards, Michael mailto:[EMAIL PROTECTED] -- Neko : One VM to run them all (http://nekovm.org)
