On 2012-06-07, Julien Ridoux <jul...@synclab.org> wrote: > On Thursday, June 7, 2012 2:26:54 AM UTC+10, unruh wrote: >> On 2012-06-06, Julien Ridoux <jul...@synclab.org> wrote: >> > On Wednesday, June 6, 2012 4:41:59 PM UTC+10, unruh wrote: >> >> On 2012-06-06, Harlan Stenn <st...@ntp.org> wrote: >> > Dear unruh (Dr. William Unruh?), >> > > >> >> > >> > I would encourage you to read the scientific papers we published. I am >> > confident you will see the level of hype being much lower. The description >> > of the algorithm you are requesting can be found in several papers. >> > Repeating myself, the main one being : ""Robust Synchronization of >> > Absolute and Difference Clocks over Networks". It does describe the >> > algorithm in more details. It may not deliver all the details you are >> > after (or be slightly outdated), but the code publicly available is also >> > there for you to read. It is the ultimate reference after all. >> > >> >> It would be nice to have a link to that publication, in particular not >> not one where you have to buy it. >> And asking people to read the code is a cop out. I have tried hard to >> read the ntpd code for example, and the chrony code. While the latter is >> slightly (only sightly) better than the former, both are a real pain to >> figure out what is going on. Reading code is always a case of far fr too >> many trees for to see the forest. >> >> For others, I have found a copy at >> http://www.cubinlab.ee.unimelb.edu.au/~darryl/Publications/synch_ToN.pdf > > Please note that the page http://www.synclab.org/docs/ I mentioned before has > links to all publications but the most recent. Please click the "Full text" > link to access the one of interest. I have updated most links for convenience > and missing ones will be added with the next update to the website. > >> > It was not my intention to describe the entire algorithm in my previous >> > message (it has been done in publicly available papers), but instead give >> > some pointers and general answers to some of the questions raised earlier. >> > My message did not intend to sparkle a heated discussion but instead try >> > to provide honest answers to this group. >> >> I am very confused by your graph comparing ntpd to radclock. The ntp has >> huge oscillations while from what I have read it is critically damped >> while what I see is a pretty high Q (of the order of 10 or so) in the >> graphs. Things I have not seen in my looks at ntpd. (Ntpd does have >> problems which I have spent time pointing out in the comparison with >> chrony) but oscillations like that I have not seen. Is this really ntpd >> or some stripped down testbed? > > I am assuming you are talking about figure 13 in the paper mentioned above. > The version of ntpd used in this plot is the stock standard version shipped > with FreeBSD 6.1 (this data set has been captured in late 2006). The machine > used is a Pentium 3 @600MHz (now long dead), a DELL Optiplex GX1 with an > onboard 3Com 100 Mbps NIC. In this experiment, ntpd is configured as a > broadcast client only listening to a stratum-1 server on the LAN. > > The broadcast configuration may be a cause for the large oscillations? I > suppose ntpd experts may have an opinion on this. > > Another paper of ours (more recent) shows more comparisons of radclock agains > ntpd under a variety of scenarios: > http://www.synclab.org/pubs/radclock_2012_TON.pdf > In particular, figures 8 to 11 show ntpd performance that may resemble what > you have observed: no erratic behaviour on the LAN if the polling period is > small enough. > > I would be interested in looking at a performance comparison of ntpd vs. > chrony if you have one accessible. I will also try to start a comparison of > chrony vs. radclock. It could be valuable to compare our findings. I have a > fair few things on my plate these days, and this may get delayed a fair bit.
There is some on my page www.theory.physics.ubc.ca/chrony/chrony.html (note that there have been changes recently so that the time tracking of all the machines right now is a mess). Miroslav Lichvar has also done extensive testing of ntpd vs chrony using a synthetic testbed (ie, the data is manufactured data but the algorithm is the same so that the system can be run at far faster than real time). I do not have the web references right now, but if you look back over the past year or so in this newsgroup for posts from him you will find the comparisons. Note, it would be really good if you persuaded your newsposting program to put in line breaks, instead of sending out huge long lines of text. Some newsreaders simply throw away the ends of lines that do not fit on one line, making reading difficult. The polling behaviour of ntpd should make no difference to the "oscillations" since the time scale of the feedback loop is adjusted to the polling period so as always to keep it roughly critically damped (modulo the feedback problems introduced by the highly non-linear clock filtering algorithm) . _______________________________________________ questions mailing list questions@lists.ntp.org http://lists.ntp.org/listinfo/questions