[LEAPSECS] NTP servers and leap seconds

Ask Bjørn Hansen ask at develooper.com
Tue Feb 1 03:17:10 EST 2011


Hi everyone,

I came across this list somewhat randomly after seeing Tony Finch's link to http://www.atnf.csiro.au/iau-comm31/activities.php ...

In the archives I saw a few postings (directly and indirectly) claiming that "for people using NTP / time sync all will be well". That's not true. NTP servers as deployed don't consistently handle it well.

Paying attention to how well NTP servers work happens to be a hobby of mine[1]. I look after a system that monitors roughly 2000 NTP servers (which in turn are used in total something like 40-100k times a second by, I think, tens of millions of clients).

After the 2008-12-31 leap second about 1 in 7 of the NTP servers went bad by just around one second in the first hour after the leap second[2]. Even 12 hours later about 2% of the servers were still bad. I didn't write it down, but I randomly checked some of the bad servers and I recall that they were using a seemingly random mix of sources (CDMA signals, GPS directly, other NTP servers over the internet etc).


- ask

[1] http://www.pool.ntp.org/
[2] https://lists.ntp.org/pipermail/pool/2009-January/004623.html

--
http://develooper.com/ - http://askask.com/


More information about the LEAPSECS mailing list