TimeSync Success Stories
That TimeSync for Windows can keep your clock synchronized with a time reference is a well-known fact. But how good is it? Here are a number of examples.
Example 1: A Compaq DeskPro
A variety of machines were used to test version 1.7 of TimeSync. A lightly loaded NT4 workstation (Compaq DeskPro) with SP4 won the prize hands down. An accurate NTP machine on the same subnet acted as the time reference. The graphs below speak for themselves. The clock was synchronized very soon after startup and the deviation was held within a narrow band. TimeSync was started 3 times on the machine during those 27 days and that is reflected in the spikes. Initially the synchronization period was 1 hour; the third time the period was increased to 4 hours.
The distribution plot of deviation shows that 73% of the deviations were in the range -1 to +1. Only approx. 9% of the deviation was outside the 5 millisecond band. This is almost ideal and, admittedly, unlikely to be surpassed.
| Deviation (milliseconds) | Distribution (in percent) |
|---|---|
| <= -5 | 3 |
| -4 | 2 |
| -3 | 3 |
| -2 | 3 |
| -1 | 23 |
| 0 | 33 |
| 1 | 17 |
| 2 | 8 |
| 3 | 2 |
| 4 | 1 |
| >= 5 | 5 |
Here are deviation plots of a few randomly chosen machines: 3k, 3o, da, de, dr, dz and e7.
We ran tests with version 1.0 on several machines and picked two particularly bad ones to demonstrate the effectiveness of the software. The two pathological cases are a Compaq Proliant server and a no-name PC both running NT 3.51 server and service pack 3. It shows, once again, that we mean it when we say: "Quality Time"!
In both cases, graphs are provided that show the clock deviation under three different conditions:
- Without TimeSync. This is what happens if you do not have any time synchronization software at all.
- With TimeSync but without adjustment enabled. This is the default mode. At every synchronization point the clock is set to the time of the time source.
- With TimeSync and adjustment. At every synchronization point the clock is speeded up or slowed down so that the deviation is corrected for. After this step TimeSync changes the clock increment to what it considers the most suitable. It determines this by watching the deviation of the immediate past.
The deviation was sampled every 5 minutes. The x-axis of the graph is time; the scale shows the sample number. The y-axis is the deviation in milliseconds.
Example 2: Compaq Proliant
These machines - at least the specimens that we found - have clocks that deviate by about 9 seconds a day. This is more than the deviation of a typical PC. I wonder what sort of clock chips were put into the machine.
The dark blue line is a straight line that shoots its way out of the graph at the rate of 9 seconds per day. Once TimeSync is started on the machine without adjustment, it looks a lot better. The saw-tooth function shows that periodically the deviation goes down to zero and the clock starts drifting away till the next synchronization point. However, if adjustment is enabled the deviation is virtually flat except in two cases: at the beginning it is trying to determine the drift constant of the clock and if the time source is not available. It requires a few periods for assessing the clocks drift. Once that is calculated the deviation stays within 50 milliseconds. Near the 500th sample, the clock begins to deviate because the time source is not available. But once the time source is back on-line the deviation is corrected.
Example 3: A no-name PC
This is a particularly bad one. The deviation is about 21 seconds per day. But TimeSync knows how to tame it.
Here again, the three curves. Once adjustment is enabled and it has been given sufficient time to calculate the drift of the clock, the deviation is almost negligible.
Additional Charts.
Once I have some more samples they will be put up here. If you have similar data send it to me and I shall try to convert it to meaningful graphs. Some statistical data may also find its way here at a later date.