Date: Sat, 27 Apr 2002 15:43:15 -0700 (PDT) From: Marcelo Carvalho <marcelomcarvalho@yahoo.com> To: freebsd-hackers@FreeBSD.org Subject: Clock granularity, dummynet & netperf Message-ID: <20020427224315.64485.qmail@web14402.mail.yahoo.com>
next in thread | raw e-mail | index | archive | help
Hi, I'm trying to use the netperf benchmark to measure the throughput for a TCP connection whose propagation delay is controlled by dummynet. I'm running these experiments on FreeBSD 4.4 machines. According to my first results (even a simple ping shows it), the delays are not properly set by dummynet. Ok, the HZ option is set initially to 100. Following Luigi's recommendations, I've increased HZ to 1000. It turns out that I'm now getting segmentation faults when I run netperf! I've tried to play around with the quantum clock as well without success... I've increased it and the measurements get really wild, totally inconsistent. I've read Luigi's previous message about the impact on select() behavior on softwares that are not aware of this. So, considering that so many people have played around with the HZ option and could perform measurements without no problem, is there a patch to netperf that overcomes this problem or a better software for performance evaluation as good (or better) than netperf??? In case not, is anybody here familiar with this problem and has fixed it in the past? I would really appreciate any help, comments, etc. Thanks, Marcelo. __________________________________________________ Do You Yahoo!? Yahoo! Health - your guide to health and wellness http://health.yahoo.com To Unsubscribe: send mail to majordomo@FreeBSD.org with "unsubscribe freebsd-hackers" in the body of the message
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20020427224315.64485.qmail>