This can be public data. Georg and I are aiming to make this a blog post. The key questions that need to be answered are: - What is the distribution of clock skew for desktop? - What is the distribution of clock skew for mobile? - Do different OSs have different clock skew distributions? - Is clock skew greater in certain parts of the world? - How does clock skew compare for different machine types? This should inform analyses as well; for example, analysis of new users can use either profile_creation_date or min(submission_date). Depending on the granularity of the analysis, one should be used over the other, but it is not apparent what the granularity is right now. For example, if the ask is per-hour, clock skew is much more likely to come into play than per-month.
One interesting question for me: Is clock skew stable per client?
This should also probably include a description of how we're gathering the clock skew data. Georg, are we guaranteed that the Date header is added immediately before sending over the wire? We're not generating that timestamp in advance, or storing it along with the ping on failed upload or anything? Feel free to point me in the right direction of the code and I can take a look.
The Date header is added immediately and is not persisted: https://dxr.mozilla.org/mozilla-central/rev/20d57b9c4183973af4af5e078dff2aec0b74f928/toolkit/components/telemetry/TelemetrySend.jsm#1117 http://searchfox.org/mozilla-central/rev/d0448c970093f94bd986a21d3a9e8c366b316eb4/mobile/android/base/java/org/mozilla/gecko/telemetry/TelemetryUploadService.java#334 Both Firefox Desktop & Fennec have this since 48 (bug 1144778, bug 1268525).
(In reply to Georg Fritzsche [:gfritzsche] from comment #3) > The Date header is added immediately and is not persisted: Lets double-check that for Fennec, but it looks like immediate usage to me.