07-05-2004 01:11 PM - edited 03-02-2019 04:50 PM
Hi,
I have been doing some lab trials researching the best setting for the advertisement_interval to our MPLS-VPN eBGP peers. The trials show somewhat inconsistent behaviour which I would like explained:
(a). In the steady state, an UPDATE is received from an iBGP neighbor, the scan process (scan-time import param) inserts the associated prefixes into the VRF. These prefixes are advertised almost instantaneously to eBGP peers. A second UPDATE is received from the iBGP peer, again the scan process inserts the attached prefixes into the VRF, and then the router waits a period before sending the prefixes to the eBGP peers. (Note. The time between the receipt of the first UPDATE and the sending of the second UPDATE is equal to the configured advertisement_interval. ~ This makes sense to me.
(b) However, with the same scenario, but less of an interval between received UPDATES, in the second case all prefixes are sent to eBGP peers in a single UPDATE approximately 7 seconds less than the configured advertisement_interval.
Therefore:
1. Should (or is) the advertisement_interval adhered to for all UPDATES sent, or can the first be sent immediately?
2. If not (1), is the minimum advertisement_interval always adhered to.
3. What triggers the counters, the receipt of the UPDATE(s) or the import into the VRF. (I suspect the former) ~ The debug shows the reset update-group(s) counting.
4. What is the correlation with the import scanner process, presumably a prefix can only be advertised once installed in the VRF.
5. If (4), then is it true to say that the maximum UPDATE time is the larger of the two : scan-time import or advertisement_interval.
6. Why the difference in behavior for scenario's (a) and (b), which effectively are the same.
(I know what value's we need set, just trying to validate that my reasoning is correct).
Cheers,
Simon
07-06-2004 03:59 AM
The min advertisement interval timers is triggered by the transmission of an update to a peer, so the behaviour you're seeing looks right to me. In the first situation, the first update starts the timer, and the second update, which arrives after the update is sent (or rather, after it is built in memory/etc), is held until the timer expires.
In the second situation, the second update is received before the first update is sent, so they are bundled together.
I would probably consider 1 second as a good starting point for working with the min adv interval.
:-)
Russ.W
07-07-2004 05:14 AM
Russ,
Thanks for the update.
:-)
Simon
Discover and save your favorite ideas. Come back to expert answers, step-by-step guides, recent topics, and more.
New here? Get started with these tips. How to use Community New member guide