Abstract:
Computer networks form an essential substrate for a variety of distributed applications, but they are expensive to build and operate. This makes it important to optimize their performance so that users can derive the most benefit at the least cost. Though most networks perform well when lightly used, problems can appear when the network load increases. Loosely speaking, congestion refers to a loss of network performance when a network is heavily loaded. Since congestive phenomena can cause data loss, large delays in data transmission, and a large variance in these delays, controlling or avoiding congestion is a critical problem in network management and design. This dissertation work presenting the implementation details for congestion control in computer networks.
Since these networks carry traffic of a single type, and the traffic behavior is well known, it is possible to avoid congestion simply by reserving enough resources at the start of each call. By limiting the total number of users, each admitted call can be guaranteed to have enough resources to achieve its performance target, and so there is no congestion. However, resources can be severely underutilized, since the resources blocked by a call, even if idle, are not available to other calls.
