[ntp:questions] Looking for clean C source code of algorithms in RFC 1305

David Woolley david at ex.djwhome.demon.co.uk.invalid
Tue Sep 30 20:38:13 UTC 2008


Micky wrote:
> The C source code for clock selection algorithms included in RFC 1305

RFC 1305 should be considered obsolete, even though it's replacement is 
still in draft.  The replacement is unlikely to help you any more, 
except that it does match the implementation of NTP that has been 
current for several years now.

I thought there was only pseudo code in RFC 1305, not C code.

> is not very clean. The inputs are not well defined, global variables
> with plenty of side effects, etc. You really have to dig through the
> whole document to make sense out of it and I do not have the time for
> it.

My gut feeling is that RFC 1305 is unnecessarily difficult to read, 
partly because it tries to be written for mathematicians, not for 
software developers.  However, I doubt that anything easier to read 
exists.  If it does, I imagince it is covered by commercial confidentiality.

> 
> I am looking for clean source code that defines clearly entry proc,
> clearly defines input to be provided on the basis of responses
> received from multiple NTP servers, and clearly defines output (e.g. a

Input is only ever received from one server at a time.  That input is 
processed in relation to stored information about earlier responses from 
that server and responses from other servers.

> response with the best time offset) to be used for clock/time update.

Note that the offset is as input to the clock discipline filter. 
Although that is given in a non-normative appendix, it is effectively 
part of the normative specification, and may well have been promoted to 
formal normative status in NTP V4.  I'm saying this, because it is a 
common misconception that ntpd steps the clock by the offset when ever 
it computes one.

> In other words, source code written by professional software
> engineers.

My experience in the software industry is that that is a very rare 
commodity.  In fact the best code is often written by unpaid volunteers, 
because they tend to most knowledgeable, not constrained by coding 
standards, and have not been promoted to management after a couple of 
years of coding.

If the reference implementation doesn't satisfy you, I suspect that any 
such code only exists behind closed doors.  But as the good 
implementations are generally based on the reference implementation, I 
doubt that any actually exist.

The, obsolete, NTP V3 reference implementation contains an unambiguous, 
real C, definition of the RFC 1305 algorithms, but I think you may find 
it less easy to understand than the RFC.

If you need to implement NTP, you should adapt the reference 
implementation, preferably the current one, not the one that matches RFC 
1305; the licensing is similar to BSD.




More information about the questions mailing list