To use MPLS CoS to full advantage in your network, the following functionality must be supported:

• Multiprotocol Label Switching (MPLS)—MPLS is the standardized label switching protocol defined by the Internet Engineering Task Force (IETF).

• Cisco Express Forwarding—Cisco Express Forwarding is an advanced Layer 3 IP switching technology that optimizes performance and scalability in networks that handle large volumes of traffic and that exhibit dynamic traffic patterns.

• Asynchronous Transfer Mode (ATM)—ATM signaling support is required if you are using ATM interfaces in your network.

  If you are using only packet interfaces in your network, ATM functionality is not needed.

• QoS features:

  • Weighted fair queueing (WFQ)—Used on non-GSR platforms, WFQ is a dynamic scheduling method that allocates bandwidth fairly to all network traffic.

    WFQ applies priorities, or weights, to traffic to classify the traffic into flows and determine how much bandwidth to allow each flow. WFQ moves interactive traffic to the front of a queue to reduce response time and fairly shares the remaining bandwidth among high-bandwidth flows.

  • Weighted random early detection (WRED)—WRED is a congestion avoidance mechanism that extends RED functionality by allowing different RED parameters to be configured per IP precedence value.

    IP precedence bits, contained in the type of service (ToS) octet in the IP packet header, are used to denote the relative importance or priority of an IP packet. WRED uses these IP precedence values to classify packets into different discard priorities or classes of service.

  • Modified deficit round robin (MDRR)—Used only on GSR platforms, MDRR is a traffic class prioritization mechanism that incorporates emission priority as a facet of quality of service. MDRR is similar in function to WFQ on non-GSR platforms.

    In MDRR, IP traffic is mapped to different classes of service queues. A group of queues is assigned to each traffic destination. On the transmit side of the platform, a group of queues is defined on a per-interface basis; on the receive side of the platform, a group of queues is defined on a per-destination basis. IP packets are then mapped to these queues, based on their IP precedence value.

    These queues are serviced on a round-robin basis, except for a queue that has been defined to run in either of two ways: strict priority mode or alternate priority mode.

    In strict priority mode, the high priority queue is serviced whenever it is not empty; this ensures the lowest possible delay for high priority traffic. In this mode, however, the possibility exists that other traffic might not be serviced for long periods of time if the high priority queue is consuming most of the available bandwidth.

    In alternate priority mode, the traffic queues are serviced in turn, alternating between the high priority queue and the remaining queues.

  • Committed access rate (CAR)—CAR is a QoS feature that limits the input or output transmission rate on an interface and classifies packets by setting the IP precedence value or the QoS group in the IP packet header.

MPLS CoS functionality enables network administrators to provide differentiated services across an MPLS network. Network administrators can satisfy a wide range of networking requirements by specifying the class of service applicable to each transmitted IP packet. Different classes of service can be established for IP packets by setting the IP precedence bit in the header of each packet.

MPLS CoS supports the following differentiated services in an MPLS network:

• Packet classification

• Congestion avoidance

• Congestion management

The table below describes the MPLS CoS services and functions.

MPLS CoS enables you to duplicate Cisco IP CoS (Layer 3) features as closely as possible in MPLS devices, including label edge switch routers (edge LSRs) and label switch routers (LSRs). MPLS CoS functions map nearly one-for-one to IP CoS functions on all types of interfaces.

The table below lists the existing legacy tag switching terms and the new, equivalent Multiprotocol Label Switching (MPLS) IETF terms used in this document and other related Cisco publications.

Label switching routers (LSRs) used at the edge of a Multiprotocol Label Switching (MPLS) network backbone are devices running MPLS software. The edge LSRs can be at the ingress or the egress of the network.

At the ingress of an MPLS network, devices process packets as follows:

1. IP packets enter the edge of the MPLS network at the edge LSR.

2. The edge LSR uses a classification mechanism such as the Modular Quality of Service Command-Line Interface (CLI) (MQC) to classify incoming IP packets and set the IP precedence value. Alternatively, IP packets can be received with the IP precedence value already set.

3. For each packet, the device performs a lookup on the IP address to determine the next-hop LSR.

4. The appropriate label is inserted into the packet, and the IP precedence bits are copied into the MPLS EXP bits in the label header.

5. The labeled packets are forwarded to the appropriate output interface for processing.

6. The packets are differentiated by class according to one of the following:

   • Drop probability—Weighted random early detection (WRED)

   • Bandwidth allocation and delay—Class-based weighted fair queueing (CBWFQ)

In either case, LSRs enforce the defined differentiation by continuing to employ WRED or CBWFQ on every ingress device.

At the egress of an MPLS network, devices process packets as follows:

1. MPLS-labeled packets enter the edge LSR from the MPLS network backbone.

2. The MPLS labels are removed and IP packets may be (re)classified.

3. For each packet, the device performs a lookup on the IP address to determine the packet’s destination and forwards the packet to the destination interface for processing.

4. The packets are differentiated by the IP precedence values and treated appropriately, depending on the WRED or CBWFQ drop probability configuration.

Label switching routers (LSRs) used at the core of a Multiprotocol Label Switching (MPLS) network are devices running MPLS software. These devices at the core of an MPLS network process packets as follows:

1. MPLS labeled packets coming from the edge devices or other core devices enter the core device.

2. A lookup is done at the core device to determine the next hop LSR.

3. An appropriate label is placed (swapped) on the packet and the MPLS EXP bits are copied.

4. The labeled packet is then forwarded to the output interface for processing.

5. The packets are differentiated by the MPLS EXP field marking and treated appropriately, depending on the weighted early random detection (WRED) and class-based weighted fair queueing (CBWFQ) configuration.

You realize the following benefits when you use MPLS CoS in a backbone consisting of IP devices running Multiprotocol Label Switching (MPLS):

• Efficient resource allocation—Weighted fair queueing (WFQ) is used to allocate bandwidth on a per-class and per-link basis, thereby guaranteeing a percentage of link bandwidth for network traffic.

• Packet differentiation—When IP packets traverse an MPLS network, packets are differentiated by mapping the IP precedence bits of the IP packets to the MPLS CoS bits in the MPLS EXP field. This mapping of bits enables the service provider to maintain end-to-end network guarantees and meet the provisions of customer service level agreements (SLAs).

• Future service enhancements—MPLS CoS provides building blocks for future service enhancements (such as virtual leased lines) by meeting bandwidth requirements.

The configuration examples are based on the sample network topology shown in the figure below.