Hi All some time ago (early July 2012) I reported on the problem I was having with a galloping DS3234 which turned out to a software problem which I have subsequently fixed. Thanks again to all who made suggestions.
At the end of those posts I stated that I would set up two DS3234 and let them free run and see what happened. I started that experiment July 20 and now I want the equipment in the experiment so this is as long as I am going to let it run.
Now this test is not to the microsecond. I am reading the time by eye on LCD displays and comparing them. However it is pretty indicative. My reference time keeper is yet another DS3234 which is corrected every 24 hours to a GPS reference so I am quite confident it isn't drifting.
The number of days since July 20 to now (14 October) is 86 days or in the vicinity of 7,400,000 seconds give or take some depending on exactly what time in the day I started and stopped. I am not saying EXACTLY that number as this whole process is indicative not exact.
So of the two DS3234 I have been running one has gained 2 seconds and the other has lost 3 seconds. Now doing the sums gives me an accuracy of +0.00002703% and -0.00004054% or +0.27PPM and -0.41PPM. The stated accuracy of the DS3234 is +/- 2PPM which is about +/- 14 seconds over the period I have been testing or about 5 seconds a month. All quite approximately in an order of magnitude sense.
In terms of the specification of one my devices is about 7 times better than the spec and the other is about 5 times better. Not bad at all.
So my conclusion is that the DS3234 is pretty good. It is significantly better than an uncompensated device. It fits my applications well and I will continue to use them.
Just to again give a disclaimer as I will be hotly annoyed by any posters flaming me that this research is not as thorough as it could be. I already know that. The tests are crude and meant to give order of magnitude results. And yes I have quoted some numbers to 4 significant places and I know already that is flawed too as the underlying data fails to be good enough to support it. Also 2 units is not a statistically significant sample.
My point is that the DS3234 seems to be a decent time-keeper and the two random examples I have both do an exemplary job.
For readers information only and anyone who decides to use this device do so at their own risk.