Downloads |
Table Of Contents
Quality of Service (QoS) Fact Sheet
Congestion Avoidance and Loss Probability
Queuing and Scheduling Traffic
Multiprotocol Label Switching (MPLS)
Specific Voice QoS Considerations
Link Fragmentation and Interleaving
Fact Sheet
Quality of Service (QoS) Fact Sheet
Quality-of-Service (QoS) has already become an issue for service providers and enterprise Wide Area Networks (WANs), which are adding more voice and video traffic to an already growing amount of data traffic. For example, mission-critical and time-sensitive traffic such as voice should receive higher QoS guarantees than less time-sensitive traffic such as file transfers or e-mail. Since most are probably faced with prohibitively high costs for high-speed WAN access, just throwing more bandwidth at the problem is not an option for many points in a network. Hence optimal utilization and differentiated use of the existing bandwidth becomes a critical issue.The Internet Protocol (IP) pervasiveness to the desktop has made it the protocol of choice for emerging, end-to-end voice and video along with data applications. Thus, the challenge for network administrators and architects has been to construct their networks to support these emerging IP-based voice, video, and data applications along with their traditional circuit-oriented applications over WANs consisting of multiple media types.
Cisco provides comprehensive end-to-end IP QoS mechanisms that meet these demands regardless of the medium(s) used to construct your WAN, such as ATM, Frame Relay, and SONET. The following briefly describes many of the tools and mechanisms used to achieve end-to-end QoS and optimal utilization of WAN bandwidth.
Building QoS Network-Wide
In building an end-to-end QoS network, the functions and mechanisms are distributed between cooperating edge/aggregation devices and core/backbone switches.Packet classification and user policies are applied at the edge of the network.
The core network backbone is then responsible for high-speed switching and transport as well as policy enforcement via the congestion control and queuing techniques discussed in this document.
QoS at the Network Edge
Packet Classification
Packet classification enables network managers to specify policies that identify network traffic that can then be partitioned into multiple priority levels or classes of service (COS). The network manager can define up to six classes using the three precedence bits in the type-of-service (TOS) field in the IP packet header.After classification, the network edge ensures that packets within a class receive appropriate service levels, such as allocated rates and loss probability. Another option often used at the edge is to apply policy routing capabilities.
Packets are classified today based on criteria including physical port, IP address, application port, protocol type, or other criteria specified by either access control lists (ACLs) or extended access lists (EACLs). The criteria used to classify traffic can be set by the network administrator, when signaled by the application, or by a proxy device via the Resource Reservation Protocol (RSVP).
RSVP allows applications with real-time traffic needs to dynamically request and reserve network resources necessary to meet their specific QoS requirements. Through proxy RSVP capabilities, Cisco routers utilize RSVP to request resources on behalf of applications that are not yet RSVP-enabled.
Access Rate Allocation
The Committed Access Rate (CAR) feature also provides bandwidth management functionality. The network manager can use CAR to designate traffic handling policies when traffic either conforms to or exceeds specified rate limits. CAR rate-limit policies can be based on physical port, Media Access Control (MAC) address, IP address, application port, or other criteria specified by ACLs or EACLs. CAR rate limits can be applied to both input and output traffic and Asynchronous Transfer Mode (ATM) and Frame Relay subinterfaces. CAR can also be utilized to specify more complex bandwidth management policies via cascaded rate limits, thus providing network managers very fine-grained network resource control.Congestion Avoidance and Loss Probability
Weighted Random Early Detection (WRED) provides network managers with a powerful congestion avoidance capability that can provide preferential treatment to the different COS levels configured. Random Early Detection (RED) works cooperatively with TCP traffic sources to maximize network throughput and capacity utilization by minimizing packet loss while avoiding congestion collapse. WRED provides preferential treatment for premium traffic classes under congestion situations by allowing network managers to specify different RED thresholds and drop/cascade policies per COS.Policy-Based Routing
Policy routing provides the network manager with the means to define customized routing paths for selected packets based on criteria, such as source address and application port, not normally considered by destination-based routing protocols. Thus, particular traffic types such as voice traffic may be sent over special routes that minimize hop counts and other delay characteristics to ensure high-quality service characteristics. Also, policy routing can be used on low-end and midrange routers to classify packets and mark the packet via the IP precedence field, enabling backbone routers to give priority treatment to voice packets when congestion occurs.QoS in the Network Core
Queuing and Scheduling Traffic
The use of queuing and scheduling mechanisms to meet specified bandwidth allocation or delay bounds applies to both the output of the edge devices as well as the network's core devices.The commonly used Weighted Fair Queuing (WFQ) mechanism segregates traffic into either multiple flows or classes and then schedules traffic on the outputs to meet specified bandwidth allocation or delay bounds. WFQ classes may be assigned either by IP precedence, application ports, IP protocol, or incoming interface identified by classification at the edge.
Per Flow Queuing, Custom Queuing, Priority Queuing, and Weighted Round Robin (WRR) and Rate Scheduling are variations on the same goals of the WFQ mechanism but offer varying degrees of granularity, jitter, and network administrator control.
To meet service-level agreements, Cisco edge routers provide generic traffic shaping as well as Frame Relay and ATM traffic shaping on their output ports into a core network. Additionally, to optimize bandwidth utilization on over-subscribed output links, Cisco edge routers can perform link fragmentation/interleaving and RTP header compression.
Frame Relay and ATM Cores
These commonly used connection-oriented WAN media offer some unique QoS types (for example, ATM has service types such as Constant Bit Rate and Unspecified Bit Rate). Hence, IP QoS mechanisms, such as WRED, CAR, and IP Precedence set by the application or during classification at the edge, should be mapped to the appropriate QoS equivalents in ATM or Frame Relay circuits.Cisco offers solutions for mapping IP COS levels to a unique ATM or Frame Relay Virtual Connections (VC). IP classes, each with their own WRED threshold, can be mapped to separate ATM VC's with different service classes to provide IP COS to ATM interworking. Or, a complete set of WRED thresholds can be mapped to one VC providing congestion feedback (such as with ATM Available Bit Rate VCs) for a reduced VC count. Also, Cisco offers a solution that allows one to assign ATM SVCs or Frame Relay PVCs to specific RSVP traffic flows.
Using the above mapping capabilities ensures that a QoS policy definition that is defined and implemented over IP routers at the edge of connection-oriented cores is not completely ignored. However, there is an even better way to ensure integration of IP QoS requirements with a network core comprised of ATM switches.
Multiprotocol Label Switching (MPLS)
The IETF has a working group called Multi-protocol Label Switching (MPLS), which is working to specify the use of label switching over many media technologies, including ATM and Packet over SONET (POS). Cisco has been a pioneer by providing a prestandard MPLS solution in tag switching, with several production deployments to date.MPLS combines the scalability of routing (routers) with the speed and traffic engineering capability of switches (such as ATM switches). Thus, in an IP+ATM MPLS network, the ATM switches run the IP routing protocols (OSPF, EIGRP, BGP, and more) for the IP traffic, and may do so in addition to the UNI/PNNI protocols that may be supporting traditional connection-oriented traffic. This improves the scalability and integration of IP services, such as QoS and VPNs, over an ATM network core.
In MPLS, COS is configured on a per-link basis, instead of on a per VC basis. This greatly simplifies provisioning of IP COS in an ATM backbone. Since both the edge routers and the ATM switches are participating directly in the IP protocol's reservation messages (RSVP) in MPLS, IP QoS requirements can be directly integrated into the ATM core. This provides more consistent and better integration of IP QoS than relying solely upon an arbitrary mapping of different QoS languages at the network's edges.
MPLS allows the assignment of different label-switched-paths (LSPs)—labels are found in the cell header of each ATM VC—to individual flows or to a collection of traffic flows, where the flows have common attributes (such as source/destination address and the same QoS requirements, etc.).
MPLS brings it all together by allowing the mapping of IP QoS to the QoS attributes of each switching node and link regardless of their type, allowing a uniform QoS policy end-to-end for IP. Since MPLS is not specific to ATM or any one technology, it is possible to use it over a wide range of WAN switches.
Specific Voice QoS Considerations
Link Fragmentation and Interleaving
Voice over packet/frame-based medium is susceptible to increased latency and jitter on slow speed links when large packets/frames are queued. Link fragmentation and interleaving in Cisco's products reduces delay and jitter on slower speed links by breaking up large datagrams and interleaving low-delay traffic with the resulting smaller packets.When you send delay-sensitive traffic such as voice over packet/frame networks it is important to have an idea of the acceptable delay (200 msec typically) when you design your network.
Voice Compression
Especially for bandwidth conservation on low-speed WAN links, Cisco implements, for example, ITU G.729 and G.729a CS-ACELP to deliver up to 24 channels of toll-quality compressed voice at 8 kbps utilizing the latest in digital signal processor technology.