[chrony-users] late-arriving hardware timestamps

[Aaron Ball <aball@xxxxxxxxxxxxxx>]

Are there things we should be doing to get Chrony 4.1 to take better
advantage of hardware timestamps?

We've been using Chrony 3.2 with a patch to get hardware timestamps
(previous thread
https://listengine.tuxfamily.org/chrony.tuxfamily.org/chrony-users/2017/06/msg00017.html)
and we're preparing to upgrade to 4.1, which doesn't need the patch.
On our test hosts with X550 NICs, we're seeing fewer hardware
timestamps under 4.1 than under 3.2.

What appears to be happening is the hardware timestamps are arriving
(update_tx_timestamp) after the sample has already been processed
(NCR_ProcessRxKnown). We've used stap to trace the relevant calls in
both versions and gather some stats. For example, two one-hour runs on
the same host, one after the other:

3.2:

7956 (99.69%) timely updated timestamps averaging -85946 ns early
7 (0.09%) late updated timestamps averaging 112148 ns late
18 (0.23%) missing timestamps

Timestamp types from measurements.log:

D H: 25
H H: 7980

vs

4.1:

7601 (94.97%) timely updated timestamps averaging -77314 ns early
403 (5.03%) late updated timestamps averaging 118251 ns late
0 (0.00%) missing timestamps

Timestamp types from measurements.log:

H H: 7350
K H: 656
K K: 2

-- 
To unsubscribe email chrony-users-request@xxxxxxxxxxxxxxxxxxxx 
with "unsubscribe" in the subject.
For help email chrony-users-request@xxxxxxxxxxxxxxxxxxxx 
with "help" in the subject.
Trouble?  Email listmaster@xxxxxxxxxxxxxxxxxxxx.