Hi all, I yesterday stumbled over behavior that I cannot explain. Mabye someone here on the list has a hint.
There is a testing tool "tcpflood", which I use to carry out the tcp based tests inside the automated testbench. That tool has evolved over time and has many options. One option is to open a large number of connections and then send data. I have seen that when I request 1,000 connections, it takes quite some while to open them (with the CPU being almost idle during that time). So far, I did not try to dig into this as the only bad effect it causes is that the testbench runs a little bit longer. Yesterday, for some other reason, I ran the very same tool, with the very same options under strace. And I was totally stunned to see that the connect() calls were now finished in almost no time. When I removed strace, it took long again (30..60 seconds, not exactly measured). If I ran it under strace, maybe a second or two. This is totally reproducible. Does anybody have a clue? Thanks, Rainer PS: if someone is interested in reproducing this (or the details): the test in question is ./tests/manytcp.sh _______________________________________________ rsyslog mailing list http://lists.adiscon.net/mailman/listinfo/rsyslog http://www.rsyslog.com
