Re: [chrony-users] Hardware timestamping in CentOS 7.4 (was: Possible to get sub-millisecond accuracy?)

[Miroslav Lichvar <mlichvar@xxxxxxxxxx>]

On Fri, Jun 23, 2017 at 12:11:36PM -0400, Chris Perl wrote:
> On Fri, Jun 23, 2017 at 11:57 AM, Miroslav Lichvar <mlichvar@xxxxxxxxxx> wrote:
> > That's an interesting idea. Do you think people would prefer SW RX+HW
> > TX timestamping over SW RX+SW TX?
> 
> "SW" in this context means kernel timestamps, right?

Yes.

> On the one hand, moving toward the hardware removes variance from the
> measurements.  But, on the other, you're essentially ensuring that
> you've introduced some asymmetry to the calculation (unless you have
> the other side configured the same way).

Right, even if the server had full HW timestamping, enabling partial
HW timestamping on a client may not improve the accuracy. For
instance, on a machine with an i350 card I see that SW RX timestamps
are about 3.6 microseconds late and SW TX timestamps are about 6.4
microseconds early, so with SW RX+HW TX timestamping the clock would
be off by about 1.8 microseconds, but with SW RX+SW TX timestamping
it would be only about 1.4 microseconds. Of course, the stability
would not be as good, especially now that chrony can handle asymmetric
jitter.

> So, I suppose it could be useful, if you're careful.

I think we can support it, at least with a new option that would
disable the filter. I actually have a patch in my queue to allow
disabling the new NTP-specific filter, which should be added in 4.13
(although there is currently no HW that supports both). I can rework
the option to select a specific filter and allow "none", which would
disable RX timestamping.

> Is it the case that the kernel changes that are coming with 7.4 are
> going to allow me to use kernel timestamps for both rx and tx?  Right
> now all my chrony instances on 7.3 are using kernel timestamps for tx
> and daemon timestamps for tx.

I don't think there will be any changes in the kernel specific to
that. It should work even with the 7.3 kernel. The chrony code just
needs to be patched to enable the support for SW timestamping without
full HW timestamping support.

-- 
Miroslav Lichvar

-- 
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.