hakanson@CSE.OGI.EDU (Marion Hakanson) (03/28/91)
>Date: Wed, 27 Mar 91 17:33:45 PST >From: (John Matzka) <matzka@sequent.com> > >> In article <9103192032.AA09380@sayshell.umd.edu> louie@SAYSHELL.UMD.EDU ("Louis A. Mamakos") writes: >. . . >> I have monitored uunet.uu.net for quite a while. It does not keep good >> time being usually some large fraction of a second off. Its NTP daemon >> seems to know this but be unable to correct the local time to less than >> 1 second accuracy. I assume this represents some problem with the >> hardware or kernel. The other systems at uunet don't seem to run NTP. >> >> On the other hand for most people who are happy with the right second it >> is probably good enough. > >uunet.uu.net is a Sequent S81 running a modified version of DYNIX 3.0. As >such, it does not support the adjtime() system call (unless support for it >has been hacked up in some way) and thus cannot adjust its clock. Note There's another possible explanation for the situation at UUNET. We also run DYNIX-3.0(.17.v3) on a Sequent S81, with adjtime(2) spliced into the kernel from 4.3bsd, complete with tickadj kernel variable, etc. Our system runs ntpd-3.4 quite nicely: ogicse 63% ntpdc cse (rem) Address (lcl) Strat Poll Reach Delay Offset Disp ========================================================================== *ansel.cse.ogi.e 129.95.40.2 2 64 376 20.0 0.0 5.0 +blake.u.washing 129.95.10.2 2 1024 377 33.0 13.0 4.0 +ulysses.CS.ORST 129.95.10.2 2 1024 377 77.0 -14.0 67.0 These dispersions and offsets are typical, and we've been running ntpd on this machine for probably eight months, under all kinds of loads. Our Sequent seems to have a pretty stable clock (both hardware and software), in my experience, although the resolution is only 10ms. Now, anyone who has one of these Sequents can tell you that the ethernet interface can't handle much of a load. The UUNET machine qualifies as a box with a very high network load, I'd say, so it seems likely that its going to suffer much more than our machine does. Which means that the UUNET machine probably drops a lot of packets, which does all kinds of fun things to an NTP session. Of course, this is more speculation. As an aside, we just got our ether interfaces upgraded. These are SCED "V46/Turbo". That's right, they put tiny turbochargers on there, driven by, uh, well.... -- Marion Hakanson Domain: hakanson@cse.ogi.edu UUCP : {hp-pcd,tektronix}!ogicse!hakanson