Suggestion for healed partitions.

[Robert Woody Woodburn <[e-mail]>]

Martha, 

If all the PGs in an AD are brought up around the same time,
say during some hardware/software upgrade, it is possible that
several components will exist for a few minutes until everybody
determines reachability.  This will result in a large number
of duplicate updates tagged with different components
throughout the route servers.  Because synthesis uses all
components, this will complicate the computation.  I propose 
a very simple mechanism for removal of extra components as
an AD heals (as alluded to in section 4.1.3, page 44).

For the implementation, a simple hold-down mechanism is probably
not enough, since updates persist for many hours, although a
hold-down of, say CMTP_NEW or so would probably be wise.

When a PG who once was the AD rep determines it is no longer
an AD rep, it should issue config and dynamic messages as follows
after having waited for the hold-down:

Configuration:

	 0               8              16              24            31
	+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
	|             AD CMP            |              SEQ              |
	+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
	|             NUM TP            |
	+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+

	AD CMP := The old AD component
	SEQ    := 0 /* can't think when it wouldn't be zero */
	NUM TP := 0

Dynamic:
	
	 0               8              16              24            31
	+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
	|             AD CMP            |              SEQ              |
	+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
	|            UNAVL VG           |             NUM PS            |
	+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+

	AD CMP   := The old AD component
	SEQ      := 0 /* can't think when it wouldn't be zero */
	UNAVL VG := 0 (although this value probably is a don't care)
	NUM PS   := 0

The appropriate action for a PG receiving such updates would be to
take the usual action of removing the old updates, storing the new ones 
(which would never be used by route synthesis), and flood as usual.

This is extraordinarily easy to implement and serves a useful function.
It also doesn't require a new message type.

Any comments?

wood y