4.2.6-series Changelog

Last update: June 28, 2022 21:06 UTC (1f97faf40)

4.2.6p5

Release Date: 2011 Dec 24

No changes from 4.2.6p5-RC3.

4.2.6p5-RC3

Release Date: 2011 Dec 8

4.2.6p5-RC2

Release Date: 2011 Nov 30

4.2.6p5-RC1

Release Date: 2011 Oct 18

4.2.6p4

Release Date: 2011 Sep 22

4.2.6p4-RC2

Release Date: 2011 Aug 04

4.2.6p4-RC1

Release Date: 2011 Jul 10

4.2.6p4-beta2

Release Date: 2011 May 25

4.2.6p4-beta1

Release Date: 2011 May 16

4.2.6p3

Release Date: 2011 Jan 03

4.2.6p3-RC12

Release Date: 2010 Dec 25

4.2.6p3-RC11

Release Date: 2010 Nov 28

4.2.6p3-RC10

Release Date: 2010 Nov 14

4.2.6p3-RC9

Release Date: 2010 Nov 10

4.2.6p3-RC8

Release Date: 2010 Oct 29

4.2.6p3-RC7

Release Date: 2010 Oct 25

4.2.6p3-RC6

Release Date: 2010 Oct 24

4.2.6p3-RC5

Release Date: 2010 Oct 22

4.2.6p3-RC4

Release Date: 2010 Oct 16

4.2.6p3-RC3

Release Date: 2010 Oct 14

4.2.6p3-RC2

Release Date: 2010 Sep 25

4.2.6p3-RC1

Release Date: 2010 Sep 18

4.2.6p3-beta1

Release Date: 2010 Sep 11

4.2.6p2

Release Date: 2010 Jul 09

4.2.6p2-RC7

Release Date: 2010 Jun 19

4.2.6p2-RC6

Release Date: 2010 Jun 12

4.2.6p2-RC5

Release Date: 2010 Jun 03

4.2.6p2-RC4

Release Date: 2010 May 19

4.2.6p2-RC3

Release Date: 2010 May 11

4.2.6p2-RC2

Release Date: 2010 Apr 27

4.2.6p2-RC1

Release Date: 2010 Apr 18

4.2.6p1

Release Date: 2010 Apr 09

4.2.6p1-RC6

Release Date: 2010 Mar 31

4.2.6p1-RC5

Release Date: 2010 Feb 09

4.2.6p1-RC4

Release Date: 2010 Feb 04

4.2.6p1-RC3

Release Date: 2010 Jan 24

4.2.6p1-RC2

Release Date: 2009 Dec 25

4.2.6p1-RC1

Release Date: 2009 Dec 20

4.2.6

Release Date: 2009 Dec 09

Docs

sntp

ntpd

refclocks

Updates and cleanup from Dave Mills

I’ve now spent eleven months of a sabbatical year - 7 days a week, 6-10 hours most days - working on NTP. I have carefully reviewed every major algorithm, examined its original design and evolution from that design. I’ve trimmed off dead code and briar patches and did zillions of tests contrived to expose evil vulnerabilities. The development article is in rather good shape and should be ready for prime time.

  1. The protostats statistics files have been very useful in exposing little twitches and turns when something hiccups, like a broken PPS signal. Most of what used to be syslog messages are now repackaged as protostats messages with optional syslog as well. These can also be sent as traps which might be handy to tiggle a beeper or celltext. These, the sysstats files and cryptostats files reveal the ambient health of a busy server, monitor traffic and error counts and spot crypto attacks.

  2. Close inspection of the clock discipline behavior at long poll intervals (36 h) showed it not doing as well as it should. I redesigned the FLL loop to improve nominal accuracy from several tens of milliseconds to something less than ten milliseconds.

  3. Autokey (again). The enhanced error checking was becoming a major pain. I found a way to toss out gobs of ugly fat code and replace the function with a much simpler and more comprehensive scheme. It resists bait-and-switch attacks and quickly detect cases when the protocol is not correctly synchronized.

  4. The interface code for the kernel PPS signal was not in sync with the kernel code itself. Some error checks were duplicated and some ineffective. I found none of the PPS-capable drivers, including the atom driver, do anything when the prefer peer fails; the kernel PPS signal remains in control. The atom driver now disables the kernel PPS when the prefer peer comes bum. This is important when the prefer peer is not a reference clock but a remote NTP server.

  5. The flake restrict bit turned out to be really interesting, especially with symmetric modes and of those especially those using Autokey. Small changes in the recovery procedures when packets are lost now avoid almost all scenarios which previously required protocol resets.

  6. I’ve always been a little uncomfortable when using the clock filter with long poll intervals because the samples become less and less correlated as the sample age exceeds the Allan intercept. Various schemes have been used over the years to cope with this fact. The latest one and the one that works the best is to use a modified sort metric where the delay is used when the age of the sample is less than the intercept and the sum of delay and dispersion above that. The net result is that, at small poll intervals the algorithm operates as a minimum filter, while at larger poll intervals it morphs to FIFO. Left unmodified, a sample could be used when twelve days old. This along with the FLL modifications has made a dramatic improvement at large poll intervals.

  7. Backward Incompatible: The state variable is no longer reported or available via ntpq output. The following system status bit names have been changed:

Other names have been changed as well. See the change history for libntp/statestr.c for more details. Other backward-incompatible changes in ntpq include:

See the change history for other details.