Re: Comments on FSM

[Alex Zinin <[e-mail]>]

Kunihiro,

> Instead of IdleHold timer we should explicitly mention about Start
> timer.  It is already implicitly used in RFC1771.  It is almost same
> as IdleHold timer but hopefully does not have impact on current
> implementation.

Or just say that implementations may use internal mechanisms
to generate the Start event and encourage people to implement
exponential holddown.

> We don't need Keep Idle flag.  Instead of that just stop Start timer.
> This mechanism can be used not only exponential backoff but also
> administratively shutdown or maximum prefix overflow.  So let's keep
> it simple.

Agreed.

> In IdleHold state there is a description for local limit of
> IdleHoldTimer.

>>         Upon entering the Idle Hold state, if the IdleHoldTimer exceeds
>>         the local limit the "Keep Idle" flag is set.

> This should be clearly defined.  For example StartTimer Ceiling.

Not sure if we even need this. See above.

> Difference between manual start and auto start is small.  I'm not sure
> separating Start event into Manual start event and Auto start event is
> good idea or not.

Yes, looks like it is the same event for FSM.
If so, it should be one event with two ways of generation.

> So my proposal is liek this.

> Status  : No new status
> Timer   : Start timer
> Variable: StartTimer
>           StartTimer Ceiling
>           Connect Retry Count

> (Possible addition
> Event   : Manual start
>           Auto start)

> IMHO, this topic needs more discussion and clarification.

>>Maybe the exponential backoff parts could be published as a separate draft
>>instead of including it in the base specification?

> I agrree with Greg.  A separate draft for exponential back off make
> sense.  I think even we have a separate draft for exponential back
> off, previous draft bgp4-16 FSM description still work fine with that.

No strong opinion on this.
If exponential back-off is optional and has been implemented
in some implementations, I think it can stay in the main spec.

Alex.