The "allocated-capacity" framework provides a controlled and predictable way to allocate bandwidth during network congestion. It supports sender-based and receiver-based methods for controlling bandwidth allocation. The framework allows for different levels of best-effort service, enabling charging for usage and more efficient network resource utilization. The paper focuses on algorithms for essential components of the framework: a differential dropping algorithm for network routers and a tagging algorithm for profile meters at the edge of the network for bulk-data transfers. Simulation results show the effectiveness of the combined algorithms in controlling TCP traffic to achieve targeted sending rates. The framework is designed to allocate bandwidth based on service allocation profiles. It allows users and providers to make capacity allocation decisions based on business and administrative goals. In the public Internet, where commercial providers offer service for payment, the framework allows providers to charge different prices to users with different service requirements. In private networks, administrative measures are often used to allocate resources. The framework provides a means to allocate different resources to different users. The mechanism provides useful information to providers about provisioning requirements. With the framework in place, service providers can more easily allocate specific levels of assured capacity to customers and can easily monitor their networks to detect when their customers' needs are not being met. The paper describes the "allocated-capacity" framework in detail, including a preferential dropping algorithm for network routers and a tagging algorithm tailored for bulk-data TCP traffic. It also presents simulation results showing the effectiveness of the framework in providing different levels of best-effort service with high assurance over the existing Internet. The framework also provides a simple way of identifying nonresponsive users at aggregation points. The paper also discusses related work, including dynamic allocation of bandwidth, priority scheduling, and weighted fair queueing. It describes the framework's approach to tagging packets as in or out and treating them differently based on the tags. The framework incorporates the idea of tagging packets and extends it in three aspects: 1) instantiates the framework by designing a set of tagging and dropping algorithms; 2) provides a simple way to identify and isolate nonresponsive connections; and 3) demonstrates the effectiveness of the framework with simulation results. The framework is designed to provide a spectrum of services, including a dedicated link model, an aggregated commitment model, and a specified and predictable throughput model for TCP streams. The framework considers three things when describing a service allocation profile: traffic specifications, geographic scope, and probability of assurance. The framework also discusses statistical assurance, which is a matter of provisioning. In the scenario, an ISP can track the amount of traffic tagged as in crossing various links over time and provide enough capacity to carry this subset of the traffic, even at times of congestion. This is how the Internet is managed today, but the addition of tags gives the ISP a better handle on how much of the traffic at any instant is "valued" traffic and how much is discretionary or opportunistic traffic for which a more relaxed attitude can be tolerated. The paper also discusses a receiver-controlled schemeThe "allocated-capacity" framework provides a controlled and predictable way to allocate bandwidth during network congestion. It supports sender-based and receiver-based methods for controlling bandwidth allocation. The framework allows for different levels of best-effort service, enabling charging for usage and more efficient network resource utilization. The paper focuses on algorithms for essential components of the framework: a differential dropping algorithm for network routers and a tagging algorithm for profile meters at the edge of the network for bulk-data transfers. Simulation results show the effectiveness of the combined algorithms in controlling TCP traffic to achieve targeted sending rates. The framework is designed to allocate bandwidth based on service allocation profiles. It allows users and providers to make capacity allocation decisions based on business and administrative goals. In the public Internet, where commercial providers offer service for payment, the framework allows providers to charge different prices to users with different service requirements. In private networks, administrative measures are often used to allocate resources. The framework provides a means to allocate different resources to different users. The mechanism provides useful information to providers about provisioning requirements. With the framework in place, service providers can more easily allocate specific levels of assured capacity to customers and can easily monitor their networks to detect when their customers' needs are not being met. The paper describes the "allocated-capacity" framework in detail, including a preferential dropping algorithm for network routers and a tagging algorithm tailored for bulk-data TCP traffic. It also presents simulation results showing the effectiveness of the framework in providing different levels of best-effort service with high assurance over the existing Internet. The framework also provides a simple way of identifying nonresponsive users at aggregation points. The paper also discusses related work, including dynamic allocation of bandwidth, priority scheduling, and weighted fair queueing. It describes the framework's approach to tagging packets as in or out and treating them differently based on the tags. The framework incorporates the idea of tagging packets and extends it in three aspects: 1) instantiates the framework by designing a set of tagging and dropping algorithms; 2) provides a simple way to identify and isolate nonresponsive connections; and 3) demonstrates the effectiveness of the framework with simulation results. The framework is designed to provide a spectrum of services, including a dedicated link model, an aggregated commitment model, and a specified and predictable throughput model for TCP streams. The framework considers three things when describing a service allocation profile: traffic specifications, geographic scope, and probability of assurance. The framework also discusses statistical assurance, which is a matter of provisioning. In the scenario, an ISP can track the amount of traffic tagged as in crossing various links over time and provide enough capacity to carry this subset of the traffic, even at times of congestion. This is how the Internet is managed today, but the addition of tags gives the ISP a better handle on how much of the traffic at any instant is "valued" traffic and how much is discretionary or opportunistic traffic for which a more relaxed attitude can be tolerated. The paper also discusses a receiver-controlled scheme