Interesting article on how Google has moved beyond NTP, with a mention of last year's leap second: http://www.wired.com/wiredenterprise/2012/11/google-spanner-time/ --Ted
Ted Cabeen <ted@cabeen.org> writes:
Interesting article on how Google has moved beyond NTP, with a mention of last year's leap second: http://www.wired.com/wiredenterprise/2012/11/google-spanner-time/
It would be interesting to have more details, since that article makes a lot of Google not using NTP and then proceeds to describe a network protocol that sounds exactly like NTP, down to how it distributes stratum zero clock sources to other systems in the same data center. So it's not clear exactly what they changed. -- Russ Allbery (rra@stanford.edu) <http://www.eyrie.org/~eagle/>
I'm no NTP expert so I'm not sure what it does differently, but the more detailed paper is at http://research.google.com/archive/spanner.html Eric On Mon, Nov 26, 2012 at 12:40 PM, Russ Allbery <rra@stanford.edu> wrote:
Ted Cabeen <ted@cabeen.org> writes:
Interesting article on how Google has moved beyond NTP, with a mention of last year's leap second: http://www.wired.com/wiredenterprise/2012/11/google-spanner-time/
It would be interesting to have more details, since that article makes a lot of Google not using NTP and then proceeds to describe a network protocol that sounds exactly like NTP, down to how it distributes stratum zero clock sources to other systems in the same data center. So it's not clear exactly what they changed.
-- Russ Allbery (rra@stanford.edu) <http://www.eyrie.org/~eagle/>
On Mon 2012-11-26T13:10:03 -0800, Eric Fischer hath writ:
I'm no NTP expert so I'm not sure what it does differently, but the more detailed paper is at http://research.google.com/archive/spanner.html
In which it looks as if google has taken on the task of creating its own global time scale (a task previously only attempted by national governments and international agencies) and that the "leap smear" is implemented as part of that time scale. -- Steve Allen <sla@ucolick.org> WGS-84 (GPS) UCO/Lick Observatory--ISB Natural Sciences II, Room 165 Lat +36.99855 1156 High Street Voice: +1 831 459 3046 Lng -122.06015 Santa Cruz, CA 95064 http://www.ucolick.org/~sla/ Hgt +250 m
As I understand it Google is running NTP, slightly modified to smear, according to the equation lie(t) = (1.0 - cos(pi * t / w)) / 2.0 where Google doesn't say what they're using for w. For more on this, please see: Pascoe C. Time, technology and leaping seconds. Google Official Blog 2011-09-15 <http://googleblog.blogspot.com/2011/09/time-technology-and-leaping-seconds.h...> As far as I know this sort of leap-second munging was first proposed by Markus Kuhn in 2005. Kuhn has compared his proposal to what Google is doing. Please see: Kuhn M. UTC with Smoothed Leap Seconds (UTC-SLS). 2005-12-14, last modified 2011-09-15 <http://www.cl.cam.ac.uk/~mgk25/time/utc-sls/>
On 2012-11-26 at 14:22 -0800, Steve Allen wrote:
On Mon 2012-11-26T13:10:03 -0800, Eric Fischer hath writ:
I'm no NTP expert so I'm not sure what it does differently, but the more detailed paper is at http://research.google.com/archive/spanner.html
In which it looks as if google has taken on the task of creating its own global time scale (a task previously only attempted by national governments and international agencies) and that the "leap smear" is implemented as part of that time scale.
Leap smear was implemented while Google used NTP; Chris Pascoe described it publicly in: http://googleblog.blogspot.com/2011/09/time-technology-and-leaping-seconds.h... A while back, I patched up the OpenNTPd sources with code from FreeBSD and wanted to get around to implementing leap seconds handling, which OpenNTPd didn't do. I didn't get time to do so, but did document my understanding of the issues and the approaches: https://github.com/syscomet/openntpd/blob/master/LeapSeconds.md Note though, as I describe in that document, smear only works in a closed community of timeservers, because of the lie introduced on the wire; for open Internet usage, something like UTC-SLS is needed instead. In all this, the closest relationship there is to time _zones_ is that there's a similar problem of distributing a file of offsets to be used by all systems handling time. Instead of the Olson TZ DB, it's the IERS leap-second jumps. -Phil
participants (6)
-
Eric Fischer -
Paul Eggert -
Phil Pennock -
Russ Allbery -
Steve Allen -
Ted Cabeen