If the clocks drift apart then NTP will slowly turn forward or back the time of the local system. Especially back is a problem because it would then mean that a request can arrive before it was sent. Making it negative. Since we're often measuring stuff in the 30-600ms timeframe even small drifts can be noticed. The solution is to rely only on the HW time pulse or time signals that use that as a basis i.e. switch off NTP.
As I said I have seen negative results in tests and was wondering where they came from. NTP makes perfect sense. But the problem isn't negative results only but skewed response times. Since clocks only adjust very little in the scheme of things it doesn't matter but it's good to be aware of the issue. Cheers Oliver -----Original Message----- From: sebb [mailto:[email protected]] Sent: Friday, 27 February 2009 2:00 p.m. To: JMeter Users List Subject: Re: Running JMeter virtualised instances On 26/02/2009, Oliver Erlewein [DATACOM] <[email protected]> wrote: > Hi Sonam, > > Thanks for that! Lots of info in there. I'll mull over it in detail now. > Have you ever run a comparison VM to real? Did you get the same results? > > That NTP issue you describe is an interesting one because I have had that on > my physical test harness too! Didn't ever think to pin this on NTP though. It > only happens once in a blue moon but I wondered where it came from. So now I > have an explanation. Thanks!!! You could also make your life easy by just > switching off ntpd before you run the test and activate it after the test > again (provided your tests don't run more than 12 hrs because then the clock > would be significantly off). > If the clock is significantly off in 12 hours then perhaps it needs to be repaired ... but so long as the clock runs evenly, one can allow for any drift by measuring the time difference at the end of the test. > Cheers Oliver > > --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
