NetNews Usenet Archive 1993 #1/NN_1993_1.iso/spool/comp/protocol/time/ntp/1188

Internet Message Format | 1993-01-13 | 1.3 KB

Path: sparky!uunet!paladin.american.edu!howland.reston.ans.net!spool.mu.edu!agate!ucbvax!UDEL.EDU!Mills
From: [e-mail]
Newsgroups: comp.protocols.time.ntp
Subject: Re:  Problems with xntp and MIPS RISC/os 5.01
Message-ID: <[e-mail]>
Date: 13 Jan 93 04:56:00 GMT
Sender: [e-mail]
Distribution: inet
Organization: The Internet
Lines: 19

Alan,

First, make sure your clock has a resolution at least in the 10-20 ms
range; if not, forget it. Then, start xntpd and watch the offsets right
after startup. If there is a large frequency offset in your CPU, you
will see a definite trend in offsets for the first few. If the difference
in (smoothed) offsets from one sample to the next is more than one
millisecond, your frequency is out of tolerance. Run your CPU for a
day without outside synchronization and, using a wristwatch and deft
fingers, try to estimate the frequency error in parts per million.
Then, tickle adjtime to compensate for it, about one part in 10,000
for each 100 ppm. You get to guess the sign - I forget. Repeat the process
until the frequency error is less than 100 ppm.

This is all the help I can give. If your frequency error is less than
100 ppm and the stock code still jumps around, you need an operating-
system guru, not a simple clockwatcher.

Dave