Path: utzoo!utgpu!watserv1!watmath!att!pacbell.com!ucsd!ucbvax!HERMES.CHPC.UTEXAS.EDU!jones
From: jones@HERMES.CHPC.UTEXAS.EDU
Newsgroups: comp.protocols.time.ntp
Subject: Re: [Nathan H. Hillery: xntpd on convex/sgi/3090?]
Message-ID: <9101022234.AA17014@hermes.chpc.utexas.edu>
Date: 2 Jan 91 22:34:57 GMT
References: <9101021547.aa24657@huey.udel.edu>
Sender: daemon@ucbvax.BERKELEY.EDU
Distribution: inet
Organization: The Internet
Lines: 35




If you are running an older convex c120`s, c130`s, and
older c200's then you have a problem.   The older
machines loose clock interrupts when they
talk to the spu because of a kernel "printf".
The two most common cases that a convex
looses time because of a printf 
the settimeofday function and nfs errors - timeouts.

The settimeofday function loose about 200 ms because
of the messages the is sent to the spu.  Which
is bad since this is larger then 128 ms which
means that xntp and ntp will always stay
unsynchronized. You can get a patch to noop the printf
in the settimeofday function.

If you have a lot of nfs mounts you will also see
about a 200 ms glitch in the time every time
you have a nfs error.

If you have a real new convex, we have a convex C220
wide body, then you may not see the above problems.

If you don't have a lot of nfs errors I would suggest
getting the settimeofday patch from convex.

You don't have to run tickadj with the "-s" option, its
not needed.  Just run tickadj with the "-A" option or
the "-a" option to set the tickadj to a smaller value then
1 ms.


Bill Jones