Definitive MPLS Network Designs [Electronic resources] نسخه متنی

Design Lessons to Be Taken from Globenet

A number of observations can be made from the design choices Globenet made:

Delivering multiple services from the same edge platforms is possible, although a number of areas, such as scale and security, require close attention.

Layer 3 MPLS VPN services may be delivered seamlessly across multiple autonomous systems, either of the same provider or of different operators cooperating closely.

Where a simple overprovisioning strategy cannot be adopted in the core because of high bandwidth cost in many regions of the world, careful network optimization techniques can be deployed. By combining traffic engineering and a rich set of QoS mechanisms in a concerted way, strong QoS differentiation can be achieved in the core. You also can achieve tight SLAs while operating at relatively high link loads and thus minimize recurring bandwidth cost.

The use of DiffServ-Aware MPLS Traffic engineering (DS-TE) allows for optimum traffic distribution in the network on a per-class basis using differentiated call admission controls and constraint-based routing. These take into account the resources actually granted to each class by DiffServ scheduling as well as per-class engineering constraints defined by the operator.

Where the end customer's intranet spans very large geographic distances such as multiple continents, bandwidth is scarce, and significant incompressible propagation delays eat up most of the delay budget for sensitive applications. A rich set of classes of service lets you address in a cost-effective way the specific requirements of various customer applications.

Where access to the Layer 3 MPLS VPN service is supported through another service provider, consistent CoS offerings may be supported for the end customer, even when the two service providers have deployed different QoS policies in their core. This is achieved through appropriate QoS mapping functions at service provider boundaries.

Layer 3 MPLS VPN services can be extended to support IPv6 in an incremental fashion. The upgrade for IPv6 support can be localized to the subset of the edge routers that actually need to offer that service. The core needs no upgrade or configuration changes.

In parts of the network where there is much more IP traffic to carry than native ATM traffic, the ATM network can be trunked over the multiservice MPLS infrastructure. This involves support of pseudowire services on the edge routers and an appropriately engineered core to guarantee appropriate QoS and reliability.

A service provider can extend its reach to other regions via other service providers' IP/MPLS networks by deploying virtual POPs (VPOPs). The VPOP model relies on the combination of various technologies. These include inter-AS MPLS Traffic Engineering with dynamic computation of shortest interprovider constrained paths, interprovider QoS, and Fast Reroute. This provides a service equivalent to any other "regular" POP, with strict SLAs in terms of QoS and network availability. It offers a cost-effective alternative to service providers that want to extend their footprint. It also allows a very fine granularity as well as rapid adjustment in terms of interconnection bandwidth offered to the VPOP.

Automatic provisioning features can be used to ease the deployment of a full mesh of MPLS Traffic Engineering LSPs with minimal configuration.

MPLS Traffic Engineering LSPs can be automatically resized by headend routers to dynamically adjust the TE LSP bandwidth to the actual traffic demand and consequently to find the shortest path for the requested bandwidth and optimize network resources.