home *** CD-ROM | disk | FTP | other *** search
- Newsgroups: comp.protocols.time.ntp
- Path: sparky!uunet!think.com!yale.edu!news.yale.edu!wardmac1.med.yale.edu!user
- From: matt@wardsgi.med.yale.edu (Matt Healy)
- Subject: Re: Bell of PA
- Message-ID: <matt-161292135748@wardmac1.med.yale.edu>
- Followup-To: comp.protocols.time.ntp
- Sender: news@news.yale.edu (USENET News System)
- Nntp-Posting-Host: wardmac1.med.yale.edu
- Organization: Yale U. - Genetics
- References: <1992Dec15.161155.4773@fccc.edu>
- Date: Wed, 16 Dec 1992 19:17:58 GMT
- Lines: 52
-
- In article <1992Dec15.161155.4773@fccc.edu>, stodola@orion.fccc.edu (Robert
- K. Stodola) wrote:
- >
- > This is probably a FAQ question, but...
- >
- > When I first put up NTP (a long time ago), I noted that Bell of PA time was
- > two seconds faster than my computers time. Today its off by the same two
- > seconds. One of them is obviously very reliably wrong (I lean towards
- > Bell). It does seem rather curious, especially since its fast rather than
- > slow, which might be explained by other types of delays.
- >
- > I know the phone company would probably tell me that in a disagreement between
- > reality and Bell, they are definitive, but if anyone has an explanation of
- > this phenomena, I'd like to hear it.
- >
- > --
- >
- > stodola@fccc.edu -- Robert K. Stodola (occasionally) speaks for himself.
-
- Welcome to the wonderful world of highly accurate time! The
- Internet document RFC 1305 -- which defines Network Time
- Protocol -- has an extensive discussion of the things that
- can go wrong with time services. Yale's campus network time
- server frequently drifts a bit, too -- right now it is off by
- about 1 second. Local phone companies can be even worse.
-
- Personally, I rely most on my el cheapo wristwatch, which
- I *know* gains about 0.093 seconds per day. I check it
- fairly frequently against the NIST clock (available on
- various frequencies with a shortwave radio, or you can
- phone 303-499-7111).
-
- About once a week I do a linear regresion of error versus
- day of the year for the past 6 months, discard all values
- that are off by more than 0.6 second, and recompute the
- regression line. This way I am confident I always *know*
- the time to within +/- 0.25 second. Theoretically computers
- "should" be able to do better than 0.25 second, but I've yet
- to encounter any that do.
-
- [for those who may be curious: It is a CASIO F-24W for which
- I paid 15 dollars ca. 5 years ago. I have never even replaced
- a battery, and it has never to my knowledge deviated significantly
- from the 0.093 second per day rate I quoted above. The
- packaging claimed the lithium battery was good for 7 years.]
-
-
- Matt Healy
- "I pretend to be a network administrator;
- the lab net pretends to work"
-
- matt@wardsgi.med.yale.edu
-