[ntp:questions] how do I lock in average frequency correction

Terje Mathisen "terje.mathisen at tmsw.no" at ntp.org
Thu Feb 16 20:53:41 UTC 2012


Dennis Ferguson wrote:
>
> On 15 Feb, 2012, at 22:53 , Terje Mathisen wrote:
>> I do like your approach. :-)
>
> Thanks.  This was arrived at when I acquired hardware which allowed the
> system clock to be measured against an external reference with an
> accuracy in the very low 10's of nanoseconds and found the clocks in
> the kernels I had were jittering by many 100's of nanoseconds all by
> themselves.

You might be interested to know that around 1994 or so, while visiting 
friends at Novell, in Provo Utah, I fixed a bug in their (net yet 
released at the time) new NetWare version, in the low-level 
high-precision timing code (used among many other things to throttle 
network transmission rates to exactly what the connection would handle).

While running on a (brand new at the time) Pentium, the code naturally 
used the RDTSC instruction, which they then scaled to give the same ~us 
resolution ticks as the original 386/486 code.

They had a DIV in the broken code, where if the server ran for a long 
enough period, on a fast enough PC, the division could overflow.

My replacement code, which was maybe 20-30 lines total (including the 
setup code) of asm, took the raw count, multiplied by a reciprocal and 
did a double-wide shift of the result.

What does this remind you of?
:-)

Terje
-- 
- <Terje.Mathisen at tmsw.no>
"almost all programming can be viewed as an exercise in caching"



More information about the questions mailing list