[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: time resolution (was Re: six-page binary format draft)
> At 04:10 PM 11/24/97 GMT, Paul Leyland wrote:
> >> > The ISO date format "yyyy-mm-dd hh:mm:ss" is valid for another 9000
> >> Yes, please use that format. And if you permit 5, 6, 7... digit years,
> >> then it has no built in obsolescence at all.
> >While we are considering the question, is the second a fine enough
> >quantum? It's not at all obvious to me that it is. If you permit (but
> >not require) the seconds field to have a decimal fraction portion we get
> >around that limitation as well. The binary format might represent the
> >fraction as, say, integer number of nanoseconds.
> I argued at one point to cut it off at 1 minute.
> I'm fond of pointing out that Rivest and Lampson did for namespaces what
> Einstein did for spacetime. It's time we pay attention to old fashioned
> Relativity here, too.
I've rather liked that statement since I first heard it and completely
agree with its appropriateness. On the other hand, let's pay attention
to what Relativity also says.
> There is no global spacetime. It is not possible to synchronize clocks.
Agreed. But what Einstein also said is that spacetime is locally flat
and it is possible to assign a global ordering to events as long as they
lie within the light cone.
In namespace, "locally flat" translates to uniformity within a machine
(where "machine" is defined as precisely or loosely as "local") and
"within the light cone" might translate to "within a NTP-locked domain".
> The closest we can come to that is with error limits, given our knowledge of
> the minimum and maximum delay in getting a time report. If there were the
> master atomic clock on the net that still doesn't give us sub-second
> resolution of time setting.
> On the other hand, we can be reasonably sure of getting clocks to agree +/-
> 30 seconds, so maybe 1 minute resolution of times would be appropriate.
Another poster has already pointed out that it is easy to lock machines
separated by megametres to milliseconds. That's only part of my
argument though. I accept that it will be difficult to synchronize
systems scattered throughout the solar system, as has also been pointed
My point is that just because we cannot solve all the solar system's
problems at one swell floop, we should not deny ourselves the ability to
address local concerns.
I can foresee circumstances where within a locally flat portion of
namespace (to mix my metaphors) I will want authorizations to be
well-ordered with sub-second resolution. OLTP has been mentioned as an
example. Countermanding or modifying of previously given orders to a
control system might be another.
In my view, permitting but not requiring a decimal fraction of seconds
field is, at the very worst, harmless. In some circumstances it would
be useful and in others vital.