On Sat, Sep 8, 2018 at 06:40 PM, Richard L King wrote:
Michael’s posts are interesting, though I must admit to being a bit baffled as to what exactly they’re showing me ... (especially if it’s using the default W32Time settings)
No, not the default settings.
The graphs show the polling of six NTP hosts (representing UTC) reporting the offset of windows system time from each. Also, the raw average of those poll offsets (and an adjusted/corrected offset of those reported offsets).?
What I found interesting was the accelerated drift when the system was under heavy loads.
You'll see that the offsets the hosts are reporting are astoundingly variable. Best I could determine from consulting with various people is that this is due to my having cable internet, and one with particularly variable response, robbing the round-trip HTP poll timing of consistency, making the results rather noisy.
So my contention is that with an up-to-date version of Windows, with the Windows Time Service correctly configured (which means changing it from the default settings), you’ll get pretty good time syncing even if you’re not close to a stratum 1 time server (mine are about 30ms and 10-15 hops away).
?
Richard
Could be.
What do you consider "pretty good time syncing"?
Fluctuating:
- +/- a minute from UTC?
- +/- a second from UTC?
- +/- 100 ms?
- +/- 10 ms?
- +/- one ms?
- tighter?
And if drift or skew changes, how often is its bringing brought in line detected and acted upon.
How accurate of a timestamp does someone want/need. Or think they're getting.
Your "contention" is interesting, but have you a done a test run against a known UTC source to
know how closely aligned your system time is to UTC, and how often it fluctuates by what amount throughout a day, or at least your trading day?
Of course, if one isn't looking at data at a fine time scale, it won't matter at all. But if one is...
Michael
I've modified one of the graphs already uploaded. I've removed the raw average and the adjusted/corrected offset and assigned colours to each host to make it easier to track the reported offsets from a single NTP host. The first poll occurs three seconds after system time is aligned to UTC. By poll #20, 1 minute into the sampling run, system time lags UTC by ~150 ms. It is crudely aligned to UTC after poll #44. The first minutes the system is under heavy load. The next a medium load. And a light load after that.
