[ntp:hackers] smearing the leap second

Miroslav Lichvar mlichvar at redhat.com
Mon Jun 22 08:09:54 UTC 2015


On Sun, Jun 21, 2015 at 11:23:29AM -0400, Brian Utterback wrote:
> What we do want is that if all software in the network were updated and leap
> smear aware, to do the right thing. Then, replacing all of the clients with
> old software, that again it would do the right thing. Then with a mix of
> servers.

> What happens if the clients have a leap second file?

They will probably make two unnecessary steps of the clock as if a
fake leap event was announced.

> Updated
> clients should be able to have a choice of actions. We can do all of this
> with an extension field.

If the clients could be updated to be leap-smear aware, they could be
updated to slew local clock on leap second and no leap smear would be
needed, right?

Leap smearing should be used only as a last resort when serving time
to clients that can't be updated. I hope this won't become a common
thing.

> Using only a refid field doesn't allow for this and
> breaks loop detection for the duration of the smear.

I think the loop detection will break only on the client side. If a
client is synchronized with a leap smearing server, the server will
still see that the client is synchronized to it and will not select
the client for its own synchronization. Is that correct?

-- 
Miroslav Lichvar


More information about the hackers mailing list