The Centre for metrology and accreditation in Finland (MIKES,
http://www.mikes.fi ) is evaluating White Rabbit for time & frequency
GM hardware is SPEC + DIO. We currently (2014 January) run
"wr-nic-v1.0-243" which has proven stable on the long-distance link.
Input 10 MHz and 1 PPS signals to grand master.
Note that the 1 PPS signal serves only to mark which rising-edge of the
10 MHz waveform is considered the inter-second boundary. We use a
Stanford Research DG645 delay generator to delay our UTC PPS pulse
appropriately (almost 1s) for the GrandMaster. Our 10 MHz input for the
GM is derived from a 10MHz BVA oscillator locked to our UTC PPS pulse
with a phase locked loop.
Please see the following papers on how we maintain UTC MIKE:
After a "mode gm" command to lock the PLL to the external 10 MHz signal
the PTP time is reset starting at 1970. If the system time is kept
roughly accurate using NTP we can use a python script that issues a
"time setsec XX" command with the integer UTC seconds.
In 2013 April/May we performed an experiment to measure the stability of
time-transfer over short (<10 km) distances.
GrandMaster locked to UTC (MIKE)
500m fiber to WR-switch, 500m fiber back to WR-Slave
Slave 1PPS output measured compared to GrandMaster input timebase
Measurement with Agilent 53230A (20ps)
MIKES has set up a White Rabbit link between two computers connected
with a 1000 (one thousand!) kilometer link.
This work is a collaboration between MIKES and CSC (http://www.csc.fi) -
IT Center for Science Ltd that operates Funet (the Finnish National
Research and Education Network).
The White Rabbit console reports a mean round-trip-time (RTT) delay of
10.4 ms with a daily variations of ca +/- 40 ns. 10.4 ms corresponds to
a ca 2000 km round trip, assuming a refractive index of 1.5. The WR
master and slave are physically spaced at a 567 km driving distance.
We are currently evaluating the accuracy of the transferred white-rabbit
timebase using a Cs-clock at the slave site, and two geodetic
dual-frequency GPS receiver at both GM and Slave sites. This experiment
shows that the White Rabbit hardware and protocols are much more
flexible than the original design goal for use at maximum distance of
ten kilometer. Also the fact that the designs are "Open", made that
MIKES could easily replace the original SFP lasers by longer reaching
Typical RTT variations over a period of ca 48 hours. Here an older
firmware version was used and the data shows a few 8ns jumps when the
RTT crosses a critical value.
We are now (December 2013) using wr-nic-v1.0-243 firmware which so far
This graph shows RTT variations over a period of one week. The average
RTT is 10.4 milliseconds with temperature changes causing up to 50 ns
drift. Any small interruptions or instabilities in the RTT are because
problems with the clock used as input to the GrandMaster node and not
related to White Rabbit.
GrandMaster (Espoo): SPEC + DIO
SFP 1000BASE-BX TX1490nm/RX1310nm (came with starter-kit)
DWDM system (Espoo-Kajaani): Coriant hiT7300, "2,5Gbps alienwave"
NTP shared memory refclock driver
By reading the WR time from shared memory (see specmem.c in spec-sw) and
writing the time to a specific place&format in shared memory (where NTP
looks) it is possible to use WR as a refclock for NTP.
This allows placing NTP servers anywhere in your WR network that
distribute WR time.
For the experimental code see
Preliminary observation suggests that running the ptp2ntpd refclock
driver keeps the system clock offset below 0.05 ms from WR time. Note
that this data was collected with "minpoll 4 maxpoll 4" settings in
ntp.conf - smoother/better performance may be achieved by tuning these
parameters. The constant negative offset may also be tuned out with an
additional "time1" parameter in ntp.conf.