Quality of Service
Quality of Service |
|
Special mechanisms are provided to perform traffic shaping on slow links, such as 28.8-kilobyte per second (KBps) modem links. On such links, large packets can occupy the link long enough to delay small audio packets that must be sent on the same link. This can cause problems with audio quality. To avoid this problem, traffic control fragments large packets at the link layer, sending only one fragment at a time. Latency-sensitive audio packets can then be inserted in between the larger packet's fragments, thus reducing audio latency and improving audio quality.
ISSLOW, Integrated Services over Slow links, is a queuing mechanism that is used to optimize slow (low-capacity) network interfaces by reducing latency. In particular, it is designed for interfaces that forward traffic to modem links, ISDN B- channels, and sub-T1 links.
A typical packet occupies a modem link for up to half a second. Other packets queued behind it can experience significant delays. Packets that are long enough in length to exceed the maximum tolerable delay for their QoS flow are fragmented before transmission through the link, so that high-priority packets can be inserted between the fragments of the larger packet, and meet the required QoS parameters for speedy transmission. Figure 9.7 illustrates a Point-to-Point Protocol (PPP) link using ISSLOW.
Figure 9.7 PPP Link with ISSLOW
For example, a PPP link is carrying best-effort traffic in addition to guaranteed service level flow "A." The PPP link capacity is 100 KBps and the average latency is 100 milliseconds (ms). The best-effort traffic is consuming the majority of the bandwidth on the link, which is starving flow A of the resources it requires to provide guaranteed service levels. In this example, flow A cannot tolerate a delay of more than 145 ms. As the best-effort traffic fills the queue, packets from flow A arrive. The first best-effort packet (10 kilobits) is fragmented into 2-kilobit packets. The 8-kilobit packet from flow A is fragmented into 2-kilobit packets as well, and the flow A fragments are inserted between the best-effort fragments in order to meet the required latency guarantees of flow A.
