A View from Adrian Farrell
| The MPLS data plane would appear to have earned its place as a useful part of the IP network, but it continues to come under attack from the IP forwarding community. It is certainly true that many of the reasons initially cited in support of MPLS (for example, forwarding lookup performance) ceased to be valid almost as soon as MPLS was invented, but the current crop of advanced IP forwarding techniques (such as IP fast reroute and Layer 3 switching) do not look as though they will make any impact on the validity of MPLS as a data plane technology. Recent industry reports suggest that the revenue generated by providers from MPLS services will "almost double" between 2004 and 2007, representing a predicted growth of 20% per annum. However, such a rate of increase probably simply reflects a continued rollout of existing MPLS technologies and the steady migration of current services to MPLS, rather than the deployment of new MPLS features. Thus, as MPLS is increasingly used to provide connectivity for VPNs, we might expect to see a corresponding dip in the revenues for IP services. Without a doubt, MPLS is gaining momentum in the core of providers' networks. The ease with which it can offer virtual connectivity services in a wide array of configurations means that it is attractive to providers who want to offer virtual private wire, pseudowire, virtual private LANs, and VPNs across the same network. We can certainly expect to see an increase in these services as the standards settle down, and pseudowire, with its ability to establish end-to-end connectivity through LDP while encapsulating any data technology, is letting providers offer new services over their MPLS packet networks. As MPLS VPNs bed in with many successful deployments, questions are being asked about how to support multicast VPNs in an MPLS environment. Initial deployments function adequately but are known not to be scalable. More work is required to develop techniques and standards to provide multicast support without placing an undue burden on any of the provider's routers, and without overloading the network with either control or data traffic. This is a rich seam to be tapped because the use of multicast IP within enterprise networks is on the increase. In fact, the whole area of multicast support in MPLS is currently under investigation by the IETF and seems like an urgent area for research to catch up with the increased demand for the transport of IP multicast. Currently, no support exists for the distribution of MPLS labels for routing-based multicast, and work will likely be developed to bring the multicast and MPLS technologies togetherperhaps by extending the PIM routing protocol to support label distribution. Work has already begun to extend MPLS traffic engineering to support point-to-multipoint flows, and this will be a useful tool in the support of MPLS multicast VPNs. Traffic-engineered MPLS (MPLS-TE), it must be said, has seen a relatively slow start, certainly compared to the routing-based MPLS achieved by LDP. However, as MPLS networks begin to carry increasing traffic loads, traffic engineering will gain in popularity to build virtual links within the network so that traffic can more easily be balanced. At the same time, MPLS-TE will be seen as providing a useful toolset for aggregating traffic flows and graded services. The fact that MPLS-TE can be used to set up DiffServ-enabled LSPs and to make better guarantees of QoS surely confirms the importance of this technology. As market pressure forces service providers into more cooperative arrangements, we will see an increase in deployment of traffic-engineered LSPs across autonomous systems. VPNs will need to span multiple networks while guaranteeing QoS, and providers will place "virtual PoPs" within their neighbor's network to offer local connectivity outside their sphere of influence. In both cases, the establishment of TE LSPs through untrusted domains will call for new technologies and rationalized peering agreements. However, MPLS-TE must improve implementation before it is thoroughly useful. In particular, the current resource reservation techniques that are based on simple statistical models need to be replaced by real admission control functions that genuinely reserve and dedicate resources to a traffic flow and that thereby can make absolute QoS guarantees. When this technology is made available, MPLS-TE will take off as a technique for high-grade delivery across core networks. One of the biggest challenges facing MPLS deployments over the next five years is the migration of MPLS-TE control planes to GMPLS. GMPLS is a protocol family that extends the MPLS-TE protocols to support multiple nonpacket technologies such as TDM and WDM, but GMPLS also continues to support packets. Moving to a GMPLS control plane will bring a large number of benefits to the packet switching network, each of which represents an opportunity to provide new services that can open revenue streams.
As can be seen from this list, many of the features that GMPLS offers enable MPLS-TE to be considered more as a transport technology. Services provided over a GMPLS packet network will be more capable of being robust and reliable. This will mean that service level agreements can offer better guarantees not just of throughput and traffic quality, but also of availability and recovery. To successfully enable reliability, protection, and recovery service within MPLS networks, we need to see increased deployments of MPLS OAM technologies. Not only will operators need to quickly diagnose faults with an array of MPLS tools, but (more importantly) the network itself needs to be capable to detect, isolate, and report problems so that it can automatically recover in the rapid times required by high-function service level agreements. |
EAN: 2147483647
Pages: 162