Quality of service (QoS) is the enabler of the Internet Protocol (IP) network convergence trend. The fundamental concept that enables IP network convergence is the conversion or tunneling of other network technologies into Transmission Control Protocol/Internet Protocol (TCP/IP) across traditional data networks. With voice, it’s the conversion of traditional time-division multiplexing (TDM) voice streams into packetized voice packets. IP networks are built on shared bandwidth. Thus IP networks must have mechanisms that “simulate” the dedicated on-time delivery associated with traditional voice and video networks.
Moving voice, video, and other time-sensitive technologies onto traditional data networks requires that the streams be delivered on time with minimal loss and consistent delay. A delay or loss of packet in a voice stream, for example, causes a break or chop in the conversation. There is no need to retransmit a voice packet if it gets lost because that packet is only relevant at one point of time in the conversation.
Traditional data networks are less concerned with on-time delivery and loss. For example, although congestion slowing down a file download might be inconvenient and annoying, the file eventually downloads intact. The higher-layer protocols take care of making sure that all the bits transfer and that the bits end up as they were transmitted. Whether it takes 10 seconds or 10 hours for the file to transfer is irrelevant (although perhaps annoying) because the higher-layer protocols ensure the file arrives intact at the other end.
QoS is a collection of tools and practices that enable data networks to mark, prioritize, and schedule traffic. Each device on a network participates in QoS; otherwise, the quality of service varies throughout a network.
Elements of QoS include the following:
Network design—. You must design and build a network to handle the amount of data and time-sensitivity required for a company’s network convergence. All of the devices in the network must support their roles in the QoS process. For example, if a company wants to enable voice across its wide-area data network, the WAN topology must have enough bandwidth to handle the calls and existing data traffic. Additionally, the routers and WAN service provider must be able to participate in the marking, prioritization, and scheduling necessary to deliver on-time reliable voice quality.
Packet marking—. The first step in the QoS process is indicating which packets have higher importance than others. When a packet comes into the network, it is assigned a number that indicates its importance. In TCP/IP networks, this number is called the type of service (ToS), and it is a value from zero to seven (zero being the lowest and seven the highest priority).
Prioritization and queuing—. Traffic is prioritized according to the ToS setting in each packet or where the packet arrives from. Queuing is the mechanism in each device that enables the network to process higher-priority packets (such as important applications or voice traffic) before lower-priority packets. Just like the platinum frequent flyers at the airport, higher-priority packets get to jump ahead of lower-priority packets. Like a bouncer in a nightclub, if it sees too many low-priority packets in line, the router must throw out some of them to let the higher-priority packets through.
Scheduling—. After the packets are ordered by priority and processed by a network device (such as a router), they must be scheduled to go out a network interface. Interfaces with limited bandwidth (such as a WAN) might have more traffic waiting to be transmitted than there is bandwidth available. Traffic-shaping mechanisms act as air-traffic controllers at an airport in that they sort outbound traffic into priority order while also making sure the lower-priority traffic isn’t starved.
Link fragmentation and interleaving—. Networks have big packets and small packets. Video and file transfer are examples of big packets, and voice is an example of small packets. Fifteen voice packets might be the same size as one file-transfer packet. If 15 voice packets must wait for one file-transfer packet to traverse a slow WAN, it might be too late for the voice traffic to be useful. This concept is called delay, and the variation in delay that occurs is called jitter. Delay and jitter are the enemies of voice and video over data networks. Link fragmentation is the process of slicing one large packet into many smaller packets. Interleaving is the process of placing important traffic in between the less-important traffic. Therefore, voice packets continue to make it to their destination on time in between the fragments of the original file-transfer packet.
Bandwidth reservation—. Also called provisioning, this process reserves certain portions of limited bandwidth for high-priority traffic. That bandwidth is always there. If the amount of low-priority traffic exceeds the bandwidth available, the network does not use the reserved bandwidth and instead discards low-priority traffic. Bandwidth reservation is a necessary component for handling time-sensitive traffic.
The combination of QoS mechanisms and effective network design ensure the success of converging voice, video, and data onto the same network.
At-A-Glance—Quality of Service
Quality of service (QoS) refers to the perceived and measured performance of a network, typically in terms of the sound quality of a voice call or the availability of critical data. If you don’t implement a QoS strategy, applications for IP telephony, videoconferencing, and mission-critical data are subject to “best-effort” (nonguaranteed) transmission. This setup can result in choppy voice or video during times of network congestion or a loss of critical data.
This figure illustrates the differences between voice and (noncritical) data.
</division><division> <title>What Are the Problems to Solve?</title>
The three parameters that define QoS are loss, delay, and delay variance (jitter). Controlling these three factors allows you to control the QoS:
Loss—. The percentage of packets dropped. In a highly available network, loss should be less than 1 percent. Voice networks should approach 0 percent loss.
Delay—. The time it takes for a packet to reach the target destination. Delay consists of fixed delay (serialization, quantization, etc.) and variable delay (network congestion). The total end-to-end delay should be less than 150 ms.
Delay variation (or jitter)—. The difference in the delay times of consecutive packets. A jitter buffer smoothes out arrival times, but the buffer’s ability to smooth out arrival times has instantaneous and total limits. Voice networks cannot have more than 30 ms of jitter.
At-A-Glance—Ensuring Packets Arrive on Time and Intact
To mitigate the effects of loss, delay, and jitter, you must ensure that the network can properly handle time and not drop sensitive packets. To achieve QoS, you must first leave room (bandwidth) for certain packets, you must identify which packets require special treatment, and you must have rules for how to treat these packets. These three steps are also referred to as provisioning, classification, and scheduling:
Provisioning—. Ensuring that the required bandwidth is available for all applications as well as for overhead traffic.
Classification—. Marking the packet with a specific priority denoting a requirement for special service from the network. Layer 2 or Layer 3 devices can do this marking at different places in the network. Typical classification schemes identify critical (voice and mission-critical data), high (video), normal (e-mail, Internet access), and low (fax, FTP) priorities.
Scheduling—. Assigning packets to one of multiple queues (based on classification) for priority treatment through the network. A good example is a commercial-airline boarding scheme. “Now boarding rows 40 through 50. First-class passengers and VIP members may board at any time.”
At-A-Glance—Traffic Engineering and LFI
In addition to the three main steps to ensure QoS, you also need some link-specific tools, such as traffic shaping and link fragmentation and interleaving (LFI), especially when routing traffic over low-bandwidth (768 Kbps or slower) links.
Traffic shaping is throttling back packet transmission rates. If there are line-speed mismatches between remote offices, the service provider connecting the offices might be forced to drop arbitrary packets going to the slower link. To prevent networks from dropping high-priority packets, an enterprise can engineer its traffic by provisioning its traffic on the slower link. Traffic engineering also allows the enterprise to decide which packets can be, or should be, dropped (low-priority packets) when instantaneous congestion occurs.
The three most common cases for traffic engineering occur in the following situations:
Line-speed mismatches
Remote-to-central-site oversubscription
Traffic bursts above the committed rate
In addition to network congestion, one of the primary contributors to both delay and jitter is serialization delay. This delay often occurs when a time-sensitive packet gets “stuck in traffic” behind a large data packet (such as FTP data). Link fragmentation is the process of breaking up large packets to allow smaller, more time-sensitive packets to proceed through the network in a timely manner. Interleaving is the processes of “weaving” the time-sensitive packets into the train of fragmented data packets.
</division>
