Multi-master scenario

[Paul Vixie]

> > it will try them all, transfer only from a master offering the best
> > currently available serial number.  this is tricky given "sequence
> > number arithmetic" and 32-bit rounding.

> Why is this tricky?

because in sequence number arithmetic, A<B and B<C does not imply A<C.
(consider A=0, B=0x80000001, and C=FFFFFFF0, for which A>C.)

> I don't believe our serial numbers have ever wrapped around a 32-bit
> number boundry... That's a lot of changes...
> 
> Of course, in theory, both masters will always be identical within a few
> seconds...

indeed, this is mostly a theoretical problem.  but if you could control
the connectivity among three slaves, and if there were three SOAs having
this property, then theoretically they might *never* converge.  this is
one of the reasons RFC 1996/2136 proscribes loops in master/slave graphs.
-- 
Paul Vixie