494 Flow Control Chap.6 type of speed matching and use the term "congestion control"for regulating the packet population within the subnetwork.We will not make this distinction in terminology;the type and objective of flow control being discussed will be clear from the context. In this chapter we describe schemes currently used for fow control,explain their advantages and limitations,and discuss potential approaches for their improvement.In the remainder of this section we identify the principal means and objectives of flow control.In Sections 6.2 and 6.3 we describe the currently most popular flow control methods;window strategies and rate control schemes.In Section 6.4 we describe flow control in some representative networks.Section 6.5 is devoted to various algorithmic aspects of rate control schemes. 6.1.1 Means of Flow Control Generally.a need for flow control arises whenever there is a constraint on the commu- nication rate between two points due to limited capacity of the communication lines or the processing hardware.Thus.a flow control scheme may be required between two users at the transport layer,between a user and an entry point of the subnet (network layer),between two nodes of the subnet(network layer),or between two gateways of an interconnected network (internet layer).We will emphasize flow control issues within the subnet,since flow control in other contexts is in most respects similar. The term "session"is used somewhat loosely in this chapter to mean any communi- cation process to which flow control is applied.Thus a session could be a virtual circuit, a group of virtual circuits (such as all virtual circuits using the same path),or the entire packet flow originating at one node and destined for another node.Often,flow control is applied independently to individual sessions,but there is a strong interaction between its effects on different sessions because the sessions share the network's resources. Note that different sessions may have radically different service requirements. For example,sessions transferring files may tolerate considerable delay but may re- quire strict error control,while voice and video sessions may have strict minimum data rate and maximum end-to-end delay requirements,but may tolerate occasional packet losses. There are many approaches to flow control,including the following: 1.Call blocking.Here a session is simply blocked from entering the network (its access request is denied).Such control is needed,for example,when the session requires a minimum guaranteed data rate that the network cannot provide due to limited uncommitted transmission capacity.A typical situation is that of the voice telephone network and,more generally,circuit switched networks,all of which use flow control of this type.However,a call blocking option is also necessary in integrated voice,video,and data networks,at least with respect to those sessions requiring guaranteed rate.In a more general view of call blocking one may admit a session only after a negotiation of some "service contract,"for example,an agreement on some service parameters for the session's input traffic (maximum rate,minimum rate,maximum burst size.priority level,etc.)494 Flow Control Chap. 6 type of speed matching and use the tenn "congestion control" for regulating the packet population within the subnetwork. We will not make this distinction in tenninology; the type and objective of flow control being discussed will be clear from the context. In this chapter we describe schemes currently used for flow control, explain their advantages and limitations, and discuss potential approaches for their improvement. In the remainder of this section we identify the principal means and objectives of flow control. In Sections 6.2 and 6.3 we describe the currently most popular flow control methods; window strategies and rate control schemes. In Section 6.4 we describe flow control in some representative networks. Section 6.5 is devoted to various algorithmic aspects of rate control schemes. 6.1.1 Means of Flow Control Generally, a need for flow control arises whenever there is a constraint on the commu￾nication rate between two points due to limited capacity of the communication lines or the processing hardware. Thus, a flow control scheme may be required between two users at the transport layer, between a user and an entry point of the subnet (network layer), between two nodes of the subnet (network layer), or between two gateways of an interconnected network (internet layer). We will emphasize flow control issues within the subnet, since flow control in other contexts is in most respects similar. The tenn "session" is used somewhat loosely in this chapter to mean any communi￾cation process to which flow control is applied. Thus a session could be a virtual circuit, a group of virtual circuits (such as all virtual circuits using the same path), or the entire packet flow originating at one node and destined for another node. Often, flow control is applied independently to individual sessions, but there is a strong interaction between its effects on different sessions because the sessions share the network's resources. Note that different sessions may have radically different service requirements. For example, sessions transferring files may tolerate considerable delay but may re￾quire strict error control, while voice and video sessions may have strict minimum data rate and maximum end-to-end delay requirements, but may tolerate occasional packet losses. There are many approaches to flow control, including the following: 1. Call blocking. Here a session is simply blocked from entering the network (its access request is denied). Such control is needed, for example, when the session requires a minimum guaranteed data rate that the network cannot provide due to limited uncommitted transmission capacity. A typical situation is that of the voice telephone network and, more generally, circuit switched networks, all of which use flow control of this type. However, a call blocking option is also necessary in integrated voice, video, and data networks, at least with respect to those sessions requiring guaranteed rate. In a more general view of call blocking one may admit a session only after a negotiation of some "service contract," for example, an agreement on some service parameters for the session's input traffic (maximum rate, minimum rate, maximum burst size, priority level, etc.)