These QoS parameters require that an application device, such as a router, classify each packet that it receives to determine the type of data being transported. Once this is done it determines where the packet is supposed to go next and puts it in the appropriate output queue. These queues are prioritized so that delay sensitive data (voice/video) can be output at a metered rate while non-delay sensitive data is held until output bandwidth is available. Packet classification (i.e. determining what a packet contains) adds additional latency in addition to that already required for basic forwarding (i.e. determining where a packet is supposed to go). As the packet moves through higher and higher bandwidth links moving through the Metro and Core subdivisions, this latency increase can become unacceptable. Core routers receive packets so fast that they just don’t have time to classify them. As bandwidth increases it would preferable if packet processing could be kept to basic forwarding. MPLS provides a solution to this problem. As mentioned earlier in the discussion of the Level II Routing portion of the Services Pyramid, MPLS is a protocol that sets up a specific path for a given sequence of packets, identified by a label inserted into each packet. These labels are inserted and removed by application devices called Label Edge Routers (LERs). As their name implies, LERs are located at the access edge of a network subdivision, usually the Metro. The relatively low bandwidth at the Metro Access Edge gives the LER enough time to do the packet classification required to properly label the packets. This label contains information indicating where the packet is supposed to go, and what type of QoS it requires. Once the packet is labeled, Label Switch Routers (LSRs) in the Metro and Core Subdivisions can quickly process the packet just by looking at its label, which takes a lot less time than doing a full packet classification and forwarding function. An added advantage of MPLS is that it is protocol agnostic allowing LERs to easily interface a variety of protocol data streams to the MPLS core network.

Assigning QoS attributes to packets does not completely solve the problem of guaranteeing a SLA over the entire transmission route. The problem is that more than one service provider network might be required to get the packet stream from its source to its destination. In this example the Network is made up of four interconnected service provider networks. These are represented by network clouds A, B, C and D. Networks A, C and D provide access to a number of subscribers. Network B belongs to a bandwidth wholesaler whose only function is to sell bandwidth to the other networks. If Subscriber 1 wants to make a VoP call to Subscriber 2, it will be easy for Service Provider A to provide the proper QoS for the packet stream because they have control over the entire path. However, if Subscriber 1 wants to make a VoP call to Subscriber 3 (connected to Service Provider D) it will not be easy to guarantee the QoS for the packet stream. The reason for this is that the call must pass through two or more other service provider networks to reach its intended destination. Each service provider is likely to offer different services. Proving that a service offered by one carrier has met the guarantees of its SLA when the associated traffic was transiting another provider's network is not possible today. Standards don’t exist for the exchange of the statistical data needed to provide this verification. New protocols are needed that define the exchange of statistics and authentication information. It must also be noted that these service provider networks usually overlap and often compete with each other for customers. This makes predicting anything in terms of the roll-out of premium services extremely problematic, since they require the providers to cooperate. In all likelihood it will be new players who will trigger the changes.

Conclusions
Service providers will continue to increase subscriber access bandwidth, with the goal of providing a single access solution for the subscriber’s voice, video and data requirements (triple play). Service providers will differentiate themselves from each other by the services they offer. However, to remain competitive, they must keep their CapEx and OpEx under control. Expense reduction will come from modifying their existing networks to carry voice and data traffic rather than building new networks, and by reducing the numbers of application devices in their network by having individual devices provide multiple services.

Eventually the existing SONET/SDH transport system in the Metro and Core subdivisions will give way to a packet centric network possibly based on 10 Gigabit Ethernet and DWDM over fiber. MPLS will be employed throughout the Metro and Core subdivisions to provide QoS support for SLAs and to enable high speed switching on high bandwidth links. However, these changes over will be gradual due to the huge investment already made in existing equipment. Intermediate solutions such as RPR will be used in the interim.