lx.ujf.cas.cz - a member server of the pool.ntp.org

NTP Stratum 2 server; System: CPU P4 2.4GHz, ntp-4.2.6p5@1.2349-o on Linux FC3
The graphs below update every 5 minutes. Previous data can be seen by clicking a plot.

Time offset:
1ms shift was applied to the actual offset before plotting.
An offset of a clock equals to its time correction, i.e. time = clock + offset.

source: 'ntpq -c rv'
lx peers and the current source of UTC
pll & peers: last 24 hours
pll & histogram: last 10 days
temperature compensation
lx case temperature
wedge scattergrams

NTP related packet traffic:

source: 'ntpdc -c iostats'
green: received packets
blue: packets sent
traffic spikes and their fine structure
the number of unique IPs seen during one year

All the graphs are ploted by MRTG. Because negative values aren't allowed, a 1ms bias is added to the actual offset to derive the offset graph above. An ideal timekeeper would therefore display a straight line at the 1ms level (see: David Taylors' page).

Configuration changes and other events

2015 Mar 8, week 9 - lx is back in the pool.
2015 Feb 23, week 8 - Pool NTP requests are temporarily redirected to our stratum 1 server.
2014 Mar 24, week 12 - Seven hours without connection to our ISP - failure of fiber-optic cable.
2014 Jan 19, week 2 - NTP version update from ntp-4.2.0a@1.1190-r running from Jan 2005 to ntp-4.2.6p5@1.2349-o.
2012 Nov 24, week 47 - Thirty-four hours of power outage due to the high voltage trafo revision.
2010 Oct 23, week 42 - Thirty-five hours of power outage due to the high voltage trafo revision.
2008 Mar 6, week 9 - Temporary change of the path to our upstream servers (see here).
2008 Mar 1, week 8 - 4h power outage.
2007 Oct 8, week 41 - The number of requests from Turk Telecom reached 80% of total requests sent to our server. This issue persists nearly one year ...
2007 Sep 22, week 38 - Stoped the monitoring of "NTP clients". The number of pool clients grows up. Although the scripts written by Wayne Schlitt are very informative and useful, their load is too high compared to the load of ntpd itself. The last MRTG snapshot of "NTP clients" is here.
2007 Sep 17, week 38 - Net speed set to 1Gbps on the pool "Manage Servers" page.
2007 Jul 21, week 29 - `Yearly' Graphs (an offset, NTP traffic and NTP clients) from December 2005 till July 2007.
2006 Nov 2, week 44 - Started an experiment with software-based temperature compensation.
2006 Oct 20, week 42 - The server room is equiped by air-conditioning.
2006 Oct 14 and 15, week 41 - Server room rebuilding and the power reconstruction, 2x10 hours power outage.
2006 Sep 5, week 36 - Temporary changes in the uplink pathes announced by our ISP. All the connections have become highly asymmetric. See the loopstats & peerstats for that period.
2006 Sep 1, week 35 - 4.7h without uplink.
2006 Aug 27, week 34 - 5.1h power outage.
2006 Jul 15, week 28 - Ten hours power outage. Unfortunately, the transition on generator power wasn't smooth. Fifteen minutes without uplink, two server restarts.
2006 Jun 6, week 23 - Linux 2.6.12 kernel recompiled with XT-PIC timer frequency HZ=100. The original HZ=1000 value seems to be too high for precise timekeeping due to a large number of the lost interrupts. In 2.6, a code has been added to watch for "lost ticks", but false positives may occure. Each time this happens, a spurious extra tick gets added in, causing the kernel's clock to go faster than real time. Consecutively, as soon as the peers are questioned, the ntpd time offset steps down by 1 ms (the source of this info).
The prefer keyword has been removed.
2006 Apr 6, week 14 - The prefer keyword is used to cut down a large amount of clock-hopping.
2006 Jan 1, week 52 - The leapsecond insertion at GMT midnight was flawless. From the kernel: "Jan 1 00:59:59 lx kernel: Clock: inserting leap second 23:59:60 UTC". To record this rare event I used a simple shell script. Actually, the kernel does step the clock one second backward at the start of the leap second, i.e. they run through 23:59:59 UTC twice.
2005 Dec 26, week 52 - Changed the time unit used in the offset script from 1ms to 1µs. The MRTG will use only integers in its history files and its usage is not advisable if all the input values are small (see retrospective changes in these yearly graphs).
2005 Dec 19, week 51 - lx becomes a public NTP Stratum 2 timeserver.
2005 Dec 8, week 49 - The value of MaxBytes in the ntp-pkt.cfg file (used from Dec 1) was not high enough. The peeks, if any, were lost. The new value adopted for AbsMax is 12000 (200 PPS). Moreover, the NTP clients monitoring was added.
2005 Dec 2, week 48 - lx becomes a member of the pool.ntp.org.
2005 Jul 7, week 27 - MRTG monitors the offset and packet traffic.

K. Sandler, January 2014