Re: can we talk about secure time?

[Florian Weimer <fw () deneb enyo de>]

* Hanno Böck:

> Is there any reason not to tell everyone to use tlsdate?

Some folks want to run their servers within a few milliseconds of each
other, and do not care so much about security or resiliency.  They may
even use their own internal time source (either fed by the local
standard time over NTP, or by GPS, CDMA or some other wireless time
service such as DCF-77 in Germany).

Reconciling this with cryptography is certainly a challenge.  On the
other hand, this does not have to be the default.

> What's the distro's take on this? afaik many ship ntp-based solutions
> by default.

NTP (as in protocol), yes, ntp (as in implementation), perhaps not.

I think most desktop-based distributions could get away with something
like tlsdate.

In contrast, servers with long-running connections and I/O polling
loops often do not react gracefully to jumps in time.  (I once
disconnected a few hundreds, if not thousands of users from an IRC
server just by setting its time correctly.)  Sure, you can avoid that
by using the appropriate kernel clock for timeout handling, but I have
the impression that the correct clock changes every couple of years.