paul@kuhub.cc.ukans.edu (03/07/91)
Running VMS 5.4-1 on both platforms with Multinet 2.2F (which contains NTP), a VAXstation 3100 (called Lane) and a VAX 9000-210 (called Kuhub) had been up for more than 8 days, and more than 1 day, respectively. As we made some hardware changes on our Internet router Internet became unreachable for about 105 minutes. Both Lane and Kuhub used exactly the same stratum 1 servers. At the end of the 105 minutes of isolation the time difference between Lane and Kuhub was approximately 2 seconds. That's approximately a relative drift of 1 part in 3000. Why am I seeing such a large drift?
Mills@udel.edu (03/08/91)
Paul, I have no idea why your VAXen wobbled, unless there are hidden gotchas in their timekeeping code. I ordinarily expect our chimers to stay within a few milliseconds per day after being synchronized for a few days. This assumes the machines are maintained in a normal room with temperature changes limited do a degree C or so. Now, if you were running Suns, I wouldn't be surprised. For sanity check, once you lose outside chime, your machines may form a daisy-chained loop which slowly (one step per minute) counts to 15, so that all machines eventually drop off and run independently. Wehn this occurs, each machine should continue to get updates from NTP to maintain the frequency adjustment. It's just possible that the software port neglected that little chore. If so, the wobble you are seeing is due to the intrinsic hardware that comes with the machine. Dave
dunigan@THDSUN.EPM.ORNL.GOV (Tom Dunigan 576-2522) (03/09/91)
you might check with TGV, we got them to make patches to their ntp to fix a bug and to tighten their local clock adjustment. (we never could get them to use a drift file for ntp startup). I don't know if they have incorporated those fixes into their distribution. Our discusions with them occured in early November, 1990.
adelman@tgv.com (Kenneth Adelman) (03/11/91)
In article <9103081605.AA04279@thdsun.EPM.ORNL.GOV>, dunigan@THDSUN.EPM.ORNL.GOV (Tom Dunigan 576-2522) writes... >you might check with TGV, we got them to make patches to their >ntp to fix a bug and to tighten their local clock adjustment. >(we never could get them to use a drift file for ntp startup). >I don't know if they have incorporated those fixes into their distribution. >Our discusions with them occured in early November, 1990. These fixes are included in V2.2 Rev E and later. Customers running earlier revisions of V2.2 can FTP the upgrade kit from the ANONYMOUS directory on TGV.COM (ECO-AF022.A). As I had told you earlier, support for the drift file will be included in MultiNet V3.0. Braindamage in the VAX C runtime support broke the UNIX code in a way which didn't seem worth trying to patch. Ken