David Woolley <[email protected]> writes: >Unruh wrote:
>> >> Ie contrary to the claim, chrony is actually smaller than ntp both in total >> memory used and in resident memory. There could be many reasons for this >The claim was that the algorithm was low memory, not that the program >was. The core of ntpd is the phase locked loop algorithm which has a >memory of less than a half dozen variable, most of which are to do with >adjusting the poll rate, whereas the core of chrony remembers a >significant number of samples. >ntpd compromises that low memory by remembering 8 samples of about 4 >variables for each source, but its historic origins are from discrete >hardware phase locked loops, where each variable cost an op amp and >capacitor. >> besides the algorithm used to correct the clocks. Note also that my tests >The time correction algorithm uses very little code and not much memory. >The program memory usage will be larger because ntpd has multiple clock >drivers. Also, where it can, ntpd locks down all of its code, to try to >avoid page faults. On recent Linux systems, at least, libc is severely >bloated. The hardware clock drivers is a valid point. However my comment was to counter to the claim that ntp was smaller than chrony. It is not, either in the total size or in the size locked down in memory. And chrony uses the same libc that ntp does. Computational memory not is NOT capacitors and resistors. It is free. So while arguments about the amount of memory required were valid in 1970, they are not now. As Microsoft has discovered, history can be a horrid clamp on the present. By the time ntp was being developed, harware capacitors as memory had long been replaced, first by ferrite beads and then by integrated transistors/capacitors, which were essentially free, both in cost of components and in labour costs to string them together. It is precisly ntp's use of a simply second order Markovian loop requireing minimal memory, and its extremely crude method of minimizing delay noise (that 8 event register) that makes its performace with respect to both accuracy and response time so bad. _______________________________________________ questions mailing list [email protected] https://lists.ntp.org/mailman/listinfo/questions
