Backup and Resiliency

Circuit failure protection requires the traffic load to be split across multiple circuits or requires an unloaded backup circuit to be in place for redundancy. The most common forms of circuit protection use redundant links and multihoming.

For the purposes of incremental bandwidth and link-level redundancy, multilink PPP (MLP) provides an effective bundling solution. MLP requires all links to be originated and terminated on the same physical customer edge (CE) and provider edge (PE) routers and requires the links to be fairly similar in bandwidth and latency characteristics. Also note that Frame Relay Form 16 (FRF.16) (when supported by the service provider) can offer the same benefit as MLP. Load redistribution during circuit outages is automatic. If applications sensitive to packet missequencing will be deployed, such as VoIP, flow-based load balancing (rather than packet-based) should be used.

Note that a flow here is a source- and destination-based construct at Layer 3. Flow-based load balancing automatically occurs at Layer 2 with MPLS or FRF.12. There is a caveat when implementing VoIP over MLP over multiple physical links. In these cases, VoIP uses Compressed Real-Time Transport Protocol (cRTP), because packet reordering may occur (over the separate physical links) because only fragmented packets are encapsulated with MLP. Therefore, we recommend that Multiclass Multilink PPP (MCMP) be used in this case.

For a backup circuit to be effective, detailed research into the physical circuit path is required. In many cases, separate physical facilities for entering the building are required to avoid the possibility that both circuits are disrupted at the same time, which is commonly called "backhoe fade."

Note

In many cases, circuit providers share physical facilities. Purchasing circuits from separate providers does not guarantee that they will be built on separate physical facilities.

Resilient protection against link and PE failure requires multihoming the CE router to different PE routers. Additionally, protection against CE router failure requires multiple CE routers homed to multiple PE routers (to the same PE routers). The CE-to-PE routing protocol and the CE and PE hardware platforms affect the multihoming implementation. If Border Gateway Protocol (BGP) 4 is used for the CE-to-PE routing protocol, BGP multipath load sharing is preferred for multihoming the CE router. If multiple CE routers are required, or other routing protocols are used on CE-to-PE links, designs using Gateway Load Balancing Protocol (GLBP) and the Globally Resilient IP (GRIP) architecture should be considered. Resiliency options depend on service provider support. Not all service providers support all options. Virtual Router Redundancy Protocol (VRRP) is a standards-based alternative to GLBP (see RFC 3768). Some service providers offer the option of contracting a port on an alternative node as part of a backup service. The port is accessed through ISDN or another switched circuit service in the event of a circuit failure on the primary node. Backup ports are sometimes offered at a reduced rate, because the primary and backup ports cannot consume bandwidth concurrently.

Survivable Remote Site Telephony (SRST) is recommended for subscriber sites with IP telephony. As enterprises extend their IP telephony deployments from central sites to remote offices, one of the factors considered vital in deployment is the ability to provide backup redundancy functions at the remote branch office. However, the size and number of these small-office sites preclude most enterprises from deploying dedicated call-processing servers, unified-messaging servers, or multiple WAN links to each site to achieve the high availability required.