Unfortunately, as mentioned in an earlier post Allan Deviation over a longer time scale tends to show a good result from any GPSDO. I am more interested in the short term stability over perhaps no more than 10 seconds as the GPSDO is used to calibrate test equipment and to hold UHF radio equipment on frequency.
I would call that "fortunately", because at least in the long run you get the exact
frequency from the Cs units in the GPS network with a cheap receiver.
In the short run, you are at the mercy of your local crystal oscillator. GPS works
only by averaging over a long time.
I use a Lucent GPS receiver, built by HP. It uses two MTI-260 double ovens,
one of them (in the lower plug-in) as a hot spare. The MTI-260 has options
that reach into Rb territory for long term stability. Unfortunately, MTI does
not tell us what HP was willing to pay for.
The row of 5 SMA connectors on the upper unit is my own invention.
(doubler from 5 to 10 MHz and CMOS level 1PPS)
Gerhard