home *** CD-ROM | disk | FTP | other *** search
- Newsgroups: comp.protocols.time.ntp
- Path: sparky!uunet!spool.mu.edu!yale.edu!ira.uka.de!fauern!fauna!faui45d!kardel
- From: kardel@faui45d.informatik.uni-erlangen.de (Frank Kardel)
- Subject: Re: xntp3, DOSYNCTODR_SUCKS
- References: <1992Nov9.214008.2068@elroy.jpl.nasa.gov>
- Message-ID: <BxLoJC.Ky1@immd4.informatik.uni-erlangen.de>
- Sender: news@immd4.informatik.uni-erlangen.de
- Organization: CSD., University of Erlangen
- Date: Thu, 12 Nov 1992 11:13:59 GMT
- Lines: 34
-
- stevo@elroy.Jpl.Nasa.Gov (Steve Groom) writes:
-
- >... In particular, when the machine comes back up
- >after being powered off, the clock seemed to be set to a reasonable value,
- >or at least I didn't notice that it was far enough off to be a problem.
- >In xntp3, this DOSYNCTODR_SUCKS thing has a comment that seems
- >to indicate that if I don't define this, the hardware clock doesn't get
- >updated by xntpd.
- True when you also use tickadj to disable SW clock setting from the HW clock.
-
- >Is this something new in v3?
- Yes.
-
- >Also, there is a comment
- >that says that I probably only need this if I'm not on a network. It seems
- >to me that if this is indeed a problem, the machine could wake up
- >with the clock far enough out of whack that xntpd would give up, making
- >this an issue for "networked" machines as well.
-
- As long as you stay with the HW clock within 1000 seconds of the time everything
- is fine. And each time xntp has to set the SW clock because it is of by
- more than 128ms the HW clock will also be set and thus stay "close" to what
- it should be. When you are away for a long time from the net or you are synched
- very nicely via xntp your HW clock may drift quite a bit. Therefore a periodic
- settimeofday(gettimofday) is done when you define DOSYNCTODR_SUCKS. This is fine
- for machines synchronized not better then 5ms.
- DOSYNCTODR_SUCKS pertains mostly to Suns.
-
- WARNING:
- DOSYNCTODR_SUCKS introduces a glitch of ~ 800us ervery hour on Suns (SS2) and is
- thus NOT USABLE for high precision hosts.
-
- --
- Frank Kardel (time@informatik.uni-erlangen.de)
-