BGP Prefix Independent Convergence
Table of Contents
Introduction
PIC improves BGP convergence during a failure by maintaining a backup path to a destination in a fast failover.
PIC
- The main concept to BGP PIC is similar to EIGRP's feasible successor route, but converge at a predictable time without the dependence on number of prefixes, hence the name prefix-independent convergence. PIC performs that same with few prefixes to few hundred thousand prefixes.
- Convergence time does not increase as the number of prefixes in the routing table increase, which means less packet loss during convergence.
- Reduces packet loss by eliminating the time packets are dropped between the time link/node is detected as down and the time BGP converges onto a new path. With a 20Gbps of traffic and you can reduce the outage from 2 minutes to 100 ms which equates to max of max of 300GB of traffic.
- Works for only these address families:
- IPv4
- IPv6
- VPNv4
- VPNv6
- PIC works at four levels to improve the convergence by installing an alternate path. It effects the following area:
- BGP - For each prefix a BGP router with PIC will calculate a secondary best path called alternate or backup.
- RIB - Backup path is installed in the Routing Information Base.
- FIB - Backup path is installed in the Forwarding Information Base (CEF and dCEF).
- MPLS -
—unorganized notes——
- Provides the capability to converge in ms instead of seconds or minutes.
- This is accomplished by maintaining a backup path in the FIB.
- PIC is in effect between the state of a BGP router detecting a link going down and the BGP convergence. It switches a MPLS path to the backup path that was already pre-calcualted.
- BGP convergence could be seconds or minutes that would cause packet loss.
- on IOS configured with "bgp additional-paths install" under the address family vpnv4.
- By default RR will on reflect that best path, so have to use different RDs or RTs.
IGP Convergence
- BGP PIC convergence relies heavily on IGP convergence. Make sure to tune OSPF of ISIS for fastest convergence for IOS. NX and XR already tuned.
- Use BFD for link failures.
- Change Carrier-delay to 0.
- Minimize the IGP routing table size. Convergence in IGP is still linear.
- Prefix Prioritization a feature that prioritizes which IGP prefixes are processes first in IGP convergence. Find out more Cisco IS-IS and OSPF prefix prioritisation - Beyond CCIE Blog and Implementing OSPF ASR 4.3 - Cisco
Main Concepts of BGP PIC
- Convergence depends on how quickly next-hop convergence based on:
- Fast failure detection like BFD.
- Fast IGP convergence.
- Tunneling - for certain scenarios on the edge, edge routers might have to tunnel/encapsulate traffic between edges.
- Alternalte/Backup path. With Router Reflectors, this might require feature to send multiple prefixes to the iBGP neighbors. These include:
- BGP Best-External.
- Different RD.
- BGP AddPath.
- Shadow RR.
Failure Scenarios
PIC Core
- Detection is with IGP, where the BGP NEXT-HOP IGP cost changes.
- Detection of next-hop is optimized with BGP Next-Hop Address Tracking feature (not sure if that feature is still used as the default trigger delay is high on IOS).
- Two possible failures:
- Core Node Failure - Either core node or edge node fails. Core, new cost to next-hop. Edge, new next-hop.
- Core Link Failure - New cost to next-hop.
PIC Edge
- One possible failure:
- Edge Link Failure - if PEs has a next-hop-self, a link failure to the CE will not trigger IGP next-hop cost change. In that case PE to PE encapsulation has to occur.
- PIC can be configured under the VPNv4, VRF IPv4 or Global routing table.
Additional Resources
BGP PIC Edge for IP and MPLS-VPN - Cisco IOS Configuration Guide
Deploying BGP Fast Convergence BGP PIC - CiscoLive presentation.
BGP Convergence in much less than a second - Nanog 40 (PIC)
By default IOS-XR has a nexthop trigger-delay set to 3000 msec for critical and 10000 msec for non-critical. Does this timer have to be changed to optimize PIC to converge on PIC Core failures?