https://issues.apache.org/bugzilla/show_bug.cgi?id=56966
--- Comment #5 from Christopher Schultz <ch...@christopherschultz.net> --- (In reply to Konstantin Kolinko from comment #1) > I see 1ms precision when running on Windows 7. I see 1ms running on Linux. > The last time when I observed 10ms was Windows XP, but Windows XP is now > End-of-life. +1 I don't have any non-virtual Windows instances available for testing, unfortunately. I don't trust real-time clocks on VMs. > Note that System.nanoTime() has caveats. It makes sense only when measuring > time intervals. It cannot be used to measure current time. > > req.getStartTime() is used as wall clock time value. It means that there has > to be another field in addition to req.getStartTime(). It also means that > there needs to be a change to the Log interface to pass a nano time value in > addition to milli time one. AccessLogValve could take its own timestamps in nanos, though the start time would be "after" req.getStartTime(). Or we could use (nanos / 1000) to get "better" resolution for the time-interval for a request. It seems like extra work for little benefit. (Though those experiencing 15ms-minimums would certainly argue that the benefit is great.) > Is there much interest in measuring times shorter than 1ms? Usually there is > an interest in requests that take a long time. +1 For resources that run reasonably faster than 15ms, one can use a Filter around them to collect metrics and aggregate total time over many requests to get a mean-request-time if that's what you ultimately want. -- 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
