VoIP Call Convergence over BGP

rharrell · ‎12-04-2007

VoIP call convergence is directly tied to IP convergence. I have read all the responses related to BGP BFD, BGP Neighbor Fall Over, and BGP Session Timeout. My questions is this: Has anyone tested or deployed a BGP solution (over MPLS) that allows voice calls to reconverge on an alternate path without failing. I am using 5s as the threshold for convergence (as a user will hang up by then in my opinion). I am working with a large call center organization (10 MILLION calls/month) and they are deploying IPCC and want to know if an IP WAN can support convergence around link/node failures. Again, I have read the recommendations, but have yet to test or deploy any of them (BFD etc). Can BGP converge in <5s safely???? How low can I go? Each Call center is dual homed to the MPLS cloud using BGP and large pipes DS3 & up.

wdrootz · ‎12-11-2007

In the case of BGP, after an update has been received, the time taken for the best path operation is a direct consequence of the BGP table size based on the layout of the network. This can potentially add many minutes to the overall service-restoration time where it is based on BGP convergence.

This dependency on BGP table size can be alleviated almost completely by the use of techniques such as prefix-independent rewrites within the MSE device's BGP implementation. Prefix-independent rewrite uses a separate table to store common adjacency information shared by multiple prefixes that are linked to it, as opposed to storing this information within every single entry in the forwarding table.

By using this approach, when an adjacency change is necessary, it is a simple matter of updating the common adjacency information once, rather than the size of the BGP table, thus keeping the convergence time a constant, regardless of the size of the network topology involved.