From Ulrich.Windl@rz.uni-regensburg.de Sun Jun 16 03:32:53 1996
From: Ulrich Windl <Ulrich.Windl@rz.uni-regensburg.de>
Organization: Universitaet Regensburg, Klinikum
To: Mills@huey.udel.edu
Date: Tue, 14 May 1996 09:26:04 +0200
Mime-Version: 1.0
Content-Length: 1157
Content-Type: text/plain; charset=US-ASCII
Content-Transfer-Encoding: 7BIT
Subject: Re:  Fix for kernel.tar
Priority: normal
X-Mailer: Pegasus Mail for Windows (v2.23)

On 13 May 96 at 10:43, Mills@huey.udel.edu wrote:

> Ulrich,
> 
> Thanks for that. Frankly, the hist.c and jitter.c and ntptime.c
> programs were hacked together from bits and pieces just for specific
> tasks. Occasionally, some kind soul such as you comes along and
> fixes some dangling particle. Our friends salute you.
> 
> The hist.c program was originally used to debug kernel mods, where
> the gettimeofday() syscall was purpose built to avoid setting time
> backward less than a second was to be avoided like snakes. The idea
> was that setting time backward more than that is not possible with
> xntpd, unless a real time step is intended.

You are right, usually the kernel clock shouldn't go backwards, but 
unfortunately some clocks do (e.g. the current Linux clock). As a lot 
of people might use the program to test their kernel clock's 
implementation, it might be good not only to detect negative time 
warps, but also record them. I've used gnuplot to draw the histogram, 
but unfortunately the negative bars are missing. So far for 
explaining. (The advantage in the current version is that I can use a 
double logarithmic scale).

Ulrich

