[LEAPSECS] Leap Sec vs Y2K

[Warner Losh]

On 12/11/2010 09:23, Paul Sheer wrote:

>>   I'm surprised by your claim that Telcos don't do NTP, [...]

> I'm sure many do.

>

> My point is that, if one's starting premise is that most systems in the

> world *require* second-accurate syncronization, this is simply untrue.

>

> In fact most work perfectly well even when they drift by minutes,

> and a significant portion of critical systems do drift by minutes

> in practice.

In practice, these systems must not be using NFS.  NFS gets crankly when 
the times drift this much.  Sure, it kinda works, but there's a number 
of weird effects.

Also, many of these systems appear to work, where "work" is defined as 
"not catastrophically fail," but a number of problems present themselves 
downstream.  My mail client sorts things in send time order.  This means 
that people whose system clock is wrong wind up out of order.  I'd call 
that failure, but it kinda works too.

So the magnitude of this "kinda works" degrades as the time 
synchronization between systems gets worse.


> This is orders of magnitude more error than any leap second.

One problem with the "kinda works" attitude is that is a barrier to 
entry for people whose systems need to work correctly to a much higher 
precision than "minutes".  The kinda works attitude means nobody bothers 
to correct aberrant behavior because it doesn't crash the system or 
programs.  When someone goes to deploy systems, it becomes clear that 
these problems represent a "death of a thousand cuts" situation.

That's one reason why you're seeing the push back from people on this 
list: many have tried to deploy systems where sub-second synchronization 
was required by the application and have run into many problems...

Warner

> -paul

>

>

>

>

>

> _______________________________________________

> LEAPSECS mailing list

> LEAPSECS at leapsecond.com

> http://six.pairlist.net/mailman/listinfo/leapsecs

>

>

>