| Young children often have to be coaxed and cajoled into relinquishing a toy when it is another child's turn to play with it. Computer users and applications, competing for limited network bandwidth and resources, are not much different. Users share networks more than ever: at home, at work, and while traveling. Whenever many people or devices share a resource, such as a network, that resource can be either overprovisioned or oversubscribed. Overprovisioning With overprovisioning, more resources are made available than can be consumed. The term describes situations where so much of a resource is available that if every subscriber requested it at the same time, there would still be plenty of it to go around. Overprovisioning requires lots of reserves and is thus usually expensive. The following are some well-known examples of overprovisioning: -
Food on a cruise— It seems there is always more than enough food to go around. -
Congressional parking spaces— No congressperson goes without a parking space (although they may be prioritized). -
AOL CD-ROMs— There may be more than enough for every computer user on the planet. Overprovisioning is generally rare, but it has probably been the dominant QoS strategy during the history of networking. If the network is slow, add more bandwidth! Oversubscribing Oversubscribing is much more common than overprovisioning because most of the time it is cheaper. The term describes situations where a resource has many users (subscribers) who don't all ask for the resource at the same time. Too many concurrent users could consume all of the resource. When too many subscribers request service simultaneously, service may be slow or even unavailable. Some examples of oversubscribing include the following: -
The clown's attention at a 3-year-old child's birthday party -
Batteries, bread, and highway lanes just before a hurricane -
Phone calls on Mother's Day in the U.S. Oversubscription can occur whenever there is a juncture with many inputs and few outputs—that is, in any situation where a bottleneck or funnel emerges. It also occurs where things quickly go from fast to slow; we have all seen this happen on highways. In networks, every juncture is a potential bottleneck. With network devices such as routers and switches, the total of all the inputs can greatly exceed the output capacity. High-speed links, such as LANs, frequently merge with low-speed links, such as WANs. Devices at network junctures use techniques such as queuing and prioritization to relieve oversubscription, but queues and patience are finite. Many more users, much more demand for bandwidth, and new mixes of network traffic make oversubscription a concern for network managers far into the future. The demand for more bandwidth and network connectivity seems unbounded. Most analysts still see 50 to 100 percent annual growth in Internet traffic in many regions of the U.S.[1] Already, most WANs are oversubscribed, and the avalanche of traffic challenges LANs and the network devices connecting them, as well. Ethernet inventor and industry sage Bob Metcalfe observes that the value of a network increases as the square of the number of users.[2] Users now connecting make the networks they are using more valuable to those who have not connected yet. Increased network resources are obviously needed for VoIP, and more will be required for other multimedia applications. Performance Requirements The users of a converged network have conflicting performance requirements and expectations of the network. In particular, most mission-critical business applications and most web transactions use the TCP protocol, whereas multimedia and voice traffic uses the RTP protocol—both on the same IP network. TCP is adaptive, rate based, and connection oriented; it behaves politely when the network is oversubscribed. TCP applications run as fast as they can, but gracefully back down when faced with congestion. RTP applications often don't get feedback because they are using a connectionless protocol. They send data in one direction with no acknowledgments. When faced with congestion, they don't back off—frames are just dropped, degrading the quality of what is received. Thus, existing TCP business traffic may be starved of bandwidth by ill-mannered multimedia traffic. Although a VoIP call consumes relatively little bandwidth and uses small packet sizes, voice applications have strict requirements for delay, jitter, and lost data, whereas other applications require high throughput or low response times. What happens when a 64-byte voice packet is stuck behind a large file transfer in a router queue? You would like for the small VoIP packet to move to the head of the queue to avoid the delay. Because networks are shared and can be oversubscribed, mechanisms are needed to protect users and applications from each other and to provide differentiation of service: QoS. | 
