|
| | NTP 4.2.7p241 - The best Windows NTP ever?
In December 2011 I made some tests with a near-to-release
version of NTP (4.2.7p238), to see how it compared with earlier versions, and how well
it worked on different versions of Windows. There is a detailed table of
events below, and graphs of the Offset, Frequency and Jitter reported by
NTP. You can roll your mouse over the captions below the graphs to select
a different graph to view. These results and my conclusions are pending
discussions with NTP experts.
I had the chance to test a slightly more recent beta
(4.2.7p241) on 2011-Dec-19, which has significantly improved how well NTP works
on some Windows systems, now providing an almost uniformly excellent
performance. My thanks to Dave Hart for his tireless work on NTP for
Windows.
Extracts from the Windows Event Viewer are zipped
here.
The changes listed for version 2.4.7.241 are:
(4.2.7p241) 2011/12/18 Released by Harlan Stenn <stenn@ntp.org>
* [Bug 2015] Overriding sys_tick should recalculate sys_precision.
* [Bug 2037] Fuzzed non-interpolated clock may decrease.
* [Bug 2068] "tos ceiling" default and cap changed to 15.
* Floor peer delay using system precision, as with jitter, reflecting inability to measure shorter intervals.
and I recall that it was the change to the calculation of
floor peer delay which resulted in the improvement on Windows Vista and later
systems. This change has been kept for subsequent NTP versions.
Overview
- 4.2.7p241 works well in all versions tested, and there is
now no need for the earlier
4.2.4p6 (273).
- PCs running Vista or Windows-7 benefit from the new
version.
- Two Windows-7 portables tested work better with LAN sync
than Wi-Fi (PCs Torvik and Ystad).
- PCs running PPS sources saw little on no noticeable change (PCs Alta, Bacchus, Feenix and Stamsund).
- One PC running Windows-XP is unaffected by the change (PC Narvik).
In terms of connection, the LAN connected PCs mostly behave
according to whether their version on Windows allows clock interpolation.
It's more of a lottery for those with a Wi-Fi connection. The performance
with 4.2.7p241 is, in alphabetical order:
- PPS synched
- PC Alta (modern, quad-core) - Win7/64, 30 µs jitter
- PC Bacchus (quite old, single core) - Win 2000, 34 µs
jitter
- PC Feenix (not modern, single core) - Win XP, ~4 µs
jitter (I double-checked this, as it seems so good)
- PC Stamsund (not modern, hyperthreaded) - Win-7/32, +20
µs jitter
- LAN connected
- PC Gemini - anomalous, always has been
- PC Hydra - Win-7/64, ~1 ms jitter
- PC Narvik - Win-XP, 15 µs jitter
- PC Torvik - Win-7/32, ~ 1 ms jitter
- PC Ystad - Win-7/32, ~ 1 ms jitter
- Wi-Fi connected
- PC Bergen - Vista, ~1.4 ms jitter
- PC Mercury - Win-7/32, ~8.5 ms jitter
- PC Puffin - Vista, ~ 1.2 ms jitter
- PC Ystad - Win-7/32, ~4 ms jitter
Questions outstanding
Why is the Wi-Fi sync worse on the Windows-7 Wi-Fi connected
portables? The two Vista Wi-Fi connected systems do significantly better,
albeit not as good as the LAN-connected machines. But it suggests that the
Windows-7 PCs should be able to do better.
Table of Events
PC Alta |
|
|
|
|
PPS sync
Win-7/64 |
Dec 10
07:11 |
NTP update
From: ntpd 4.2.7p214-o Sep 21 2:47:14.56 (UTC-00:00) 2011 (1)
To: ntpd 4.2.7p238-o Dec 10 4:46:28.43 (UTC-00:00) 2011 (1) |
No noticeable changes |
Jitter 20-60 µs, average 30 µs.
Offset +/-70 µs. |
| |
Dec 19
09:59 |
To: ntpd 4.2.7p241-o Dec 18 22:57:49.91 (UTC-00:00) 2011 (1) |
No noticeable changes |
Jitter 22-36 µs, average 30 µs.
Offset +/- 70 µs. |
| |
|
|
|
|
PC Bacchus |
|
|
|
|
PPS sync
Win-2000 |
Dec 10
07:29 |
NTP update
From: ntpd 4.2.7p214-o Sep 21 2:47:14.56 (UTC-00:00) 2011 (1)
To: ntpd 4.2.7p238-o Dec 10 4:46:28.43 (UTC-00:00) 2011 (1)
|
No noticeable changes |
Jitter 15-60 µs, average 36 µs. |
| |
Dec 19
07:33 |
To: ntpd 4.2.7p241-o Dec 18 22:57:49.91 (UTC-00:00) 2011 (1) |
No noticeable changes. |
Jitter 28-42 µs, average 34 µs. |
| |
|
|
|
|
PC Bergen |
|
|
|
|
Wi-Fi sync
Vista |
Dec 10
14:55 |
NTP update
From: ntpd 4.2.4p6@DLH-QPC-o May 30 3:58:32.88 (UTC) 2009 (273)
To: ntpd 4.2.7p238-o Dec 10 4:46:28.43 (UTC-00:00) 2011 (1)
|
Offsets became wild and jitter increased
a lot. |
Offsets became wild, increasing from
about +/- 1.5 ms to +7/-10 ms.
Jitter increased from ~1 ms to ~6 ms. |
| |
15:25 |
Set NTPD_USE_INTERP_DANGEROUS=1 |
Slight reduction in jitter? |
Jitter dropped to a maximum of ~4-5 ms. |
| |
16:41 |
Set NTPD_USE_SYSTEM_CLOCK=1 |
Jitter increase. |
Jitter back to 5-6-7 ms. |
| |
19:11 |
Reverted to: ntpd 4.2.4p6@DLH-QPC-o May 30 3:58:32.88 (UTC) 2009 (273) |
|
Jitter and offset reverted to previous
values. Jitter just over 1 ms average, offset about +/- 1.5 ms. |
| |
Dec 20
14:23 |
To: ntpd 4.2.7p241-o Dec 18 22:57:49.91 (UTC-00:00) 2011 (1) |
No noticeable change from the previous
"good" version. |
Jitter 1.2-1.7 ms, average ~1.4ms.
Offset reduced slightly to +/- 0.9 ms, but with occasional spikes to +/- 2
ms. |
| |
|
|
|
|
| PC Feenix |
|
|
|
|
PPS sync
Win-XP |
Dec 10
07:47 |
NTP update
From: ntpd 4.2.7p213-o Sep 12 13:29:25.65 (UTC-00:00) 2011 (1)
To: ntpd 4.2.7p238-o Dec 10 4:46:28.43 (UTC-00:00) 2011 (1) |
No noticeable effect |
Jitter average varied between 2.5 and 4.5 µs.
Offset +/- 80 µs |
| |
Dec 20
11:20 |
To: ntpd 4.2.7p241-o Dec 18 22:57:49.91 (UTC-00:00) 2011 (1) |
No significant effect. |
The averaged jitter may have increased
slightly from 3.5 µs to 3.7 µs. Offset +/- 60 µs but over a much
more limited period. |
| |
|
|
|
|
| PC Gemini |
|
|
|
|
LAN sync
Win Vista |
Dec 10
07:23 |
NTP update
From: ntpd 4.2.7p213@1.2677-o Sep 20 15:44:02.43 (UTC-00:00) 2011 (1)
To: ntpd 4.2.7p238-o Dec 10 4:46:28.43 (UTC-00:00) 2011 (1) |
|
Jitter increased from a peaky 1-7 ms, to a
variable 4-9 ms.
Offset changed from a about 1 ms "noise" with positive-only
peaks up to 37 ms, to tens of milliseconds of a lower frequency
"noise" with peaks to +50/-17 ms. |
| |
15:23 |
Set NTPD_USE_INTERP_DANGEROUS=1 |
|
Rather short to be sure, but jitter slightly
worse. |
| |
16:44 |
Set NTPD_USE_SYSTEM_CLOCK=1 |
|
Jitter possibly worse at 8-18 ms.
Peaky? |
| |
19:19 |
Reverted to ntpd 4.2.7p214-o Sep 21 2:47:14.56 (UTC-00:00) 2011 (1) |
Previous performance not restored. Why
not? |
Jitter 6-13 ms. |
| |
09:15 |
Removed NTPD_USE_SYSTEM_CLOCK and restarted
NTP. |
|
|
| |
Dec 19
17:51 |
To: ntpd 4.2.7p241-o Dec 18 22:57:49.91 (UTC-00:00) 2011 (1) |
|
Averaged jitter much reduced, from ~7.5 ms
to ~2.5 ms, but still quite peaky. Offset considerably different;
periods of near-steady values with peaks. The peaks were likely
present before, but not as visible as the offset was noisier. |
| |
|
|
|
|
| PC Hydra |
|
|
|
|
LAN sync
Win-7/64 |
Dec 10
07:34 |
NTP update
From: ntpd 4.2.7p237@1.2714-o Dec 06 9:25:31.60 (UTC-00:00) 2011 (4)
To: ntpd 4.2.7p238-o Dec 10 4:46:28.43 (UTC-00:00) 2011 (1) |
No change detected. |
Average jitter 4-6ms, offset variable with
asymmetry, +14/-8 ms. Hints of a 17-minute periodicity, but also
with a two sine-wave interference pattern. |
| |
20:23 |
Restored to:
ntpd 4.2.4p6@DLH-QPC-o May 30 3:58:32.88 (UTC) 2009 (273) |
Much better. |
Jitter drop to 1.1 ms. Offset now a
random +/- ~1ms. |
| |
Dec 19
18:25 |
To: ntpd 4.2.7p241-o Dec 18 22:57:49.91 (UTC-00:00) 2011 (1) |
Even better than (273). |
Averaged jitter under 1 ms. RMS offset
dropped from ~0.5 ms to ~0.3 ms. |
| |
|
|
|
|
| PC Mercury |
|
|
|
|
Wi-Fi sync
Win-7/32 |
Dec 20 |
NTP Update
From: ntpd 4.2.4p6@DLH-QPC-o May 30 3:58:32.88 (UTC) 2009 (273)
To: ntpd 4.2.7p241-o Dec 18 22:57:49.91 (UTC-00:00) 2011 (1) |
Very similar to (273). |
Averaged jitter perhaps dropped a little,
from 9.5 to 8.5 ms, but still stabilising. Offset changed to a coarser (lower-frequency)
pattern, with greater excursions (typically was
+13/-7 ms, now +20/-5 ms). |
| |
Dec 21
14:00 |
Tried a direct LAN connection. |
(Not plotted) |
The averaged jitter increased from ~7 ms to
just over 10 ms. Offset changed to a much higher frequency pattern,
with +6/-11 ms typical limits. Surprised that jitter became worse! |
| |
|
|
|
|
| PC Molde |
|
|
|
|
Wi-Fi sync
Win-7/64 |
Dec 19
17:55 |
NTP Update
From: ntpd 4.2.4p6@DLH-QPC-o May 30 3:58:32.88 (UTC) 2009 (273)
To: ntpd 4.2.7p241-o Dec 18 22:57:49.91 (UTC-00:00) 2011 (1) |
Noted that precision had been determined as
2^-21s, hence the restart. |
Averaged jitter down from ~2 ms to often 1
ms. RMS offset reduced from ~1.5 ms to ~0.5 ms, but still quite
variable with transients. |
| |
18:11 |
Set NTPD_USE_SYSTEM_CLOCK=1 and restarted
NTP. |
Improvement. |
|
| |
|
|
|
|
| PC Narvik |
|
|
|
|
LAN sync
Win-XP |
Dec 10
07:52 |
NTP update
From: ntpd 4.2.7p214-o Sep 21 2:47:14.56 (UTC-00:00) 2011 (1)
To: ntpd 4.2.7p238-o Dec 10 4:46:28.43 (UTC-00:00) 2011 (1) |
No change noted. |
Averaged jitter 10..30 µs. Offset
+320/-210 µs, random. |
| |
Dec 19
07:17 |
To: ntpd 4.2.7p241-o Dec 18 22:57:49.91 (UTC-00:00) 2011 (1) |
No significant change. |
Averaged jitter 10-26 µs. ~15 µs
average. Offset
+350/-210 µs. |
| |
|
|
|
|
| PC Puffin |
|
|
|
|
Wi-Fi sync
Vista |
Dec 10
07:43 |
NTP update
From: ntpd 4.2.7p237@1.2714-o Dec 06 4:28:35.69 (UTC-00:00) 2011 (3)
To: ntpd 4.2.7p238-o Dec 10 4:46:28.43 (UTC-00:00) 2011 (1) |
Slight jitter increase. |
Averaged jitter increased from ~5.5 ms to
~7 ms. Elements of 10 minute periodicity in the offset before and
after the change. |
| |
Dec 19
10:50 |
To: ntpd 4.2.7p241-o Dec 18 22:57:49.91 (UTC-00:00) 2011 (1) |
Great improvement. |
Averaged jitter dropped from 6.5 ms to 1.1
ms. Jitter slightly higher at 1.25 averaged after 24 hours.
Offset now much smaller and more "noise" like. RMS offset
dropped from ~5 ms or more to ~0.5 ms, increasing to ~0.7 ms after ~24
hours. |
| |
|
|
|
|
| PC Stamsund |
|
|
|
|
PPS sync
Win-7/32 |
Dec 10
07:17 |
NTP update
From: ntpd 4.2.7p214-o Sep 21 2:47:14.56 (UTC-00:00) 2011 (1)
To: ntpd 4.2.7p238-o Dec 10 4:46:28.43 (UTC-00:00) 2011 (1) |
No obvious change? |
Averaged jitter 15..25
µs. Immediately after the change both offset and jitter seemed
slightly less, but returned to the pre-change values after 10-12 hours. |
| |
Dec 20
11:17 |
To: ntpd 4.2.7p241-o Dec 18 22:57:49.91 (UTC-00:00) 2011 (1) |
No obvious change. |
Averaged jitter 16..23 µs. RMS offset
may have increased slight from 13 µs to 15 µs. |
| |
|
|
|
|
| PC Torvik |
|
|
|
|
Wi-Fi sync
LAN-sync
Win-7/32 |
Dec 10
|
NTP update
From: ntpd 4.2.7p206
To: ntpd 4.2.7p238-o Dec 10 4:46:28.43 (UTC-00:00) 2011 (1) |
Impossible to see any change as the Wi-Fi keeps disconnecting,
making NTP's job difficult, if not impossible! |
|
| |
Dec 19
09:00 |
Changed to LAN connection. |
|
|
| |
10:36 |
NTP update
From: ntpd 4.2.7p238-o Dec 10 4:46:28.43 (UTC-00:00) 2011 (1)
To: ntpd 4.2.7p241-o Dec 18 22:57:49.91 (UTC-00:00) 2011 (1)
|
|
Averaged jitter went from something in the
7.5 ms region, down to just under 1 ms. Offset changed from peaks of
+/- 3 ms, to a steadier value with an RMS of ~0.25 ms, but also with a
period component of about one cycle per 22 minutes. |
| |
|
|
|
|
| PC Ystad |
|
|
|
|
Wi-Fi sync
Win-7/32 |
Dec 10
05:52 |
NTP update PC
From: (ntpd 4.2.7p237 test version)
To: ntpd 4.2.7p238-o Dec 10 4:46:28.43 (UTC-00:00) 2011 (1) |
No obvious change. |
Averaged jitter 3-7 ms. Offset is
asymmetric +9/-13 ms. |
| |
Dec 13
09:17 |
Restored to:
ntpd 4.2.4p6@DLH-QPC-o May 30 3:58:32.88 (UTC) 2009 (273) |
Much better |
Averaged jitter dropped from 5-5.5 ms down to ~1.3 ms.
Offset had a much higher frequency appearance, of a much lower amplitude (4.8 ms RMS dropped to ~1.0 ms RMS). |
| |
Dec 19
07:46 |
To: ntpd 4.2.7p238-o Dec 10 4:46:28.43 (UTC-00:00) 2011 (1), for start of
test run. |
|
Averaged jitter 1.5 - 6.5 ms, very
variable. Offset very spiky, +8/-10 ms. |
| |
11:22 |
To: ntpd 4.2.7p241-o Dec 18 22:57:49.91 (UTC-00:00) 2011 (1) |
|
|
| |
Dec 20
09:55 |
Turned off IPv6 on the adapters. |
No change. |
|
| |
10:25 |
Rebooted with just a LAN connection. |
Much better. |
Averaged jitter just over 1 ms. Offset
reduced to mostly + 1.4/-0.7 ms, around 0.4 ms RMS. |
| |
Dec 21
14:35 |
Rebooted with just the Wi-Fi connection. |
Little change. Eh?
(not plotted). |
Averaged jitter just over 1.1 ms, initially,
but it increased to about 5 ms after about 50 minutes, and looks more
stable there. Offset became increasingly wild, up to at least +/-3
ms. |
Graphs
2011-Dec-10
Summary: Dec-10 at 07:11 update from p214 to p238
produced no noticeable change.
2011-Dec-19
Summary: Dec-19 at 09:59 update from p238 to p241
produced no noticeable change.
2011-Dec-10
Summary: Dec-10 07:29 updated from p214 to p238,
no significant change.
2011-Dec-19
Summary: Dec-19 07:33 updated from p238 to p241,
no significant change.
2011-Dec-10
Summary: Dec-10 14:55 from 4.2.4p6(273) to
4.2.7p238 resulted in much poorer performance, so reverted at 19:11.
2011-Dec-20
Summary: Dec-20 14:23 from 4.2.4p6(273) to
4.2.7p241. No significant change, previous good performance retsined.
2011-Dec-10
Summary: Dec-10 07:47 from 4.2.7p213 to p238 - no
noticeable change.
2011-Dec-20
Summary: Dec-20 11:20 from 4.2.7p238 to p241 - no
noticeable change.
2011-Dec-10
Summary: Dec-10, 4.2.7p213 to p238 update at
07:23. Reverted to p214 at 19:19, but unable to restore previous
performance?
2011-Dec-19
Summary: Dec-19 at 17:51, update from p214 to
p241, and previous performance restored. Offset graph now noticeably
different.
2011-Dec-10
Summary: Dec-10 07:34 from p237 to p238, no
change. 20:23 back to 4.2.4p6(273) to restore good performance.
2011-Dec-19
Summary: Dec-19 18:25 from 4.2.4p6(273) to 4.2.7p241.
Improvements in jitter and RMS offset.
2011-Dec-20
Summary: Dec-19 07:52 from 4.2.4p6(273) to 4.2.7p241.
Slight improvement, but look at the offset and frequency graphs.
2011-Dec-19
Summary: Dec-19 07:52 from 4.2.7p214 to 4.2.7p241. Noticeable improvement.
2011-Dec-10
Summary: Dec-10 07:52 from 4.2.7p214 to 4.2.7p238. No significant change.
2011-Dec-20
Summary: Dec-19 07:17 from 4.2.7p238 to 4.2.7p241. No significant change.
2011-Dec-10
Summary: Dec-10 07:43 from 4.2.7p237 to 4.2.7p238.
2011-Dec-19
Summary: Dec-19 07:43 from 4.2.7p238 to 4.2.7p241. Great improvement.
2011-Dec-10
Summary: Dec-10 07:17 from 4.2.7p214 to 4.2.7p238.
2011-Dec-20
Summary: Dec-20 11:17 from 4.2.7p238 to 4.2.7p241.
Summary: Dec-19 10:36 from 4.2.7p238 to 4.2.7p241.
2011-Dec-10
Summary: Dec-10 05:52 from 4.2.7p237 to 4.2.7p238.
2011-Dec-19-20
Summary: Dec-19, 07:46 switched to 4.2.7p238, 11:22 to
4.2.7p241, Dec-20 10:25 changed from Wi-Fi to LAN connection and got even better
performance than when running on p238.
|
