BGP Finite State Machine Timing Question

[kfarrington]

If a BGP neighbor is configured or is reset, how longon does it take to go from the following steps :-

Idle,

connect

active

opensend

openconfirm

established,

The reason, I ask, is that it takes up to 40 seconds on a 2600 router but on a 6500 it is much quicker. Also, I seem to see iBGP peers startup and get to the established state much quicker than eBGP peers.

This obvisously has an effect and needs carfeul consideration whilst designing BGP networks as if you have time sensitive data on a FR network and your lose the BGP peering, fast failover or the manipulation of the BGP timers is very important, but you need to set the timers to the lowest value for reconvergence.

Many thx

[kfarrington]

More info

Important to note that when you do a "clear ip bgp *" the neighbor takes an extra "estab-idle-to-active time" to startup than if doing a "cle ip bgp " where it goes from estab-idle-active" immediately. Why?

Also, note that the "open active, delay" is bigger for eBGP peers than iBGP peers. Why?

What actually is the "open active delay" and is it variable?

Also, it takes longer on a 2600 than a 6500 MSFC to get to the established state.

These are important questions to understand from a design viewpoint.

:)

[DALE FRANCIS]

Dude,

The clearing * neigbours will cause BGP to clear all tables and an amount of time randomization would be included to stop neigbour sync, if you had say 20 neigbours in a RR group and 5 EBGP neigbours and all came up at once you would have all neigbours coming up at once and posting updates at once etc hence creating potential high CPU etc...

If you clear a single neigbour then its just routes etc associated with that neigbour.

This is my view and is not fact based on Cisco IOS etc, it just seems logical.