Re: bgp4-17 Cease subcode

[Jeffrey Haas <[e-mail]>]

Gentlefolk,

Perhaps this will add some clarification on "what does this mean to me"
to have the "new state" in the FSM:

1. RFC 1771 has exponential backoff documented.
1a. The current draft allows a ceiling on this value.
2. The original FSM documents 6 states.  When you attempt to expand
   the state table from the normative text (even without the
   distinction of a manual vs. automatic start/stop events), you arrive
   at another undocumented state.
3. The draft explicitly documents this state and makes explicit the
   difference between the manual vs. the automatic events.  The
   state table the Sue had posted some time ago (wasn't that supposed
   to be in the appendix?) covers all the events and the states.
4. What does this do to implementations that supported the exponential
   backoff?  Probably not much.  You can LIVE with the existing Idle
   state for both MIB and for CLI display purposes.  What was originally
   implied was that there was a timer (you could show it in your CLI
   for example) that would implement the exponential backoff.
5. The new draft makes the state explicitly visible.  As this primarily
   affects the display within a CLI and the MIB rather than the on-the-wire
   protocol, there's no need to panic.

Implementation:
If you implement exponential backoff, it should be pretty trivial to
insert the new "state" into you FSM to deal with the un-expired timer
event.  It will be 7 in the MIB obviously. :-)

If we don't have enough wiggle room in the spec to deal with those
who don't implment the expoential backoff, perhaps a SHOULD can be
added in the appropriate spot - or the spec should remain silent
on the issue and let people be non-compliant.

Sue, Yakov: Is the fact that the new state would only be visible by
a outside of protocol scope means going to stop us from going to
full standard?  Technically, it is measurable by watching for
the packets and the delays between them but this doesn't break
interoperability.

-- 
Jeff Haas 
NextHop Technologies