500 Flow Control Chap.6 6.2 WINDOW FLOW CONTROL In this section we describe the most frequently used class of flow control methods.In Sections 6.2.1 to 6.2.3,the main emphasis is on flow control within the communication subnet.Flow control outside the subnet,at the transport layer,is discussed briefly in Section 6.2.4. A session between a transmitter A and a receiver B is said to be window'flow controlled if there is an upper bound on the number of data units that have been trans- mitted by A and are not yet known by A to have been received by B(see Fig.6.5).The upper bound (a positive integer)is called the window size or,simply,the window.The transmitter and receiver can be,for example.two nodes of the communication subnet,a user's machine and the entry node of the communication subnet,or the users'machines at the opposite ends of a session.Finally,the data units in a window can be messages, packets,or bytes,for example. The receiver B notifies the transmitter A that it has disposed of a data unit by sending a special message to A,which is called a permit (other names in the literature are acknowledgment,allocate message,etc.).Upon receiving a permit,A is free to send one more data unit to B.Thus,a permit may be viewed as a form of passport that a data unit must obtain before entering the logical communication channel between A and B.The number of permits in use should not exceed the window size. Permits are either contained in special control packets.or are piggybacked on regular data packets.They can be implemented in a number of ways:see the practical examples of Section 6.4 and the following discussion.Note also that a window flow control scheme for a given session may be combined with an error control scheme for the session,where the permits also play the role of acknowledgments;see Section 2.8.2 and the descriptions of the ARPANET and the Codex network in Section 6.4. The general idea in the window strategy is that the input rate of the transmitter is reduced when permits return slowly.Therefore,if there is congestion along the communication path of the session.the attendant large delays of the permits cause a natural slowdown of the transmitter's data rate.However,the window strategy has an additional dimension,whereby the receiver may intentionally delay permits to restrict DU DU Transmitter Receiver Permit Permit Total number of data units and permits window size WAs Figure 6.5 Window flow control between a transmitter and a receiver consists of an upper bound on the number of data units and permits in transit inside the network.500 6.2 WINDOW FLOW CONTROL Flow Control Chap. 6 In this section we describe the most frequently used class of flow control methods. In Sections 6.2.1 to 6.2.3, the main emphasis is on flow control within the communication subnet. Flow control outside the subnet, at the transport layer, is discussed briefly in Section 6.2.4. A session between a transmitter A and a receiver B is said to be window flow controlled if there is an upper bound on the number of data units that have been trans￾mitted by A and are not yet known by A to have been received by B (see Fig. 6.5). The upper bound (a positive integer) is called the window size or, simply, the window. The transmitter and receiver can be, for example, two nodes of the communication subnet, a user's machine and the entry node of the communication subnet, or the users' machines at the opposite ends of a session. Finally, the data units in a window can be messages, packets, or bytes, for example. The receiver B notifies the transmitter A that it has disposed of a data unit by sending a special message to A, which is called a permit (other names in the literature are acknowledgment, allocate message, etc.). Upon receiving a permit, A is free to send one more data unit to B. Thus, a permit may be viewed as a form of passport that a data unit must obtain before entering the logical communication channel between A and B. The number of permits in use should not exceed the window size. Permits are either contained in special control packets, or are piggybacked on regular data packets. They can be implemented in a number of ways; see the practical examples of Section 6.4 and the following discussion. Note also that a window flow control scheme for a given session may be combined with an error control scheme for the session, where the permits also play the role of acknowledgments; see Section 2.8.2 and the descriptions of the ARPANET and the Codex network in Section 6.4. The general idea in the window strategy is that the input rate of the transmitter is reduced when permits return slowly. Therefore, if there is congestion along the communication path of the session, the attendant large delays of the permits cause a natural slowdown of the transmitter's data rate. However, the window strategy has an additional dimension, whereby the receiver may intentionally delay permits to restrict Transmitter Receiver A B ~~ Total number of data units and permits <; window size WAS Figure 6.5 Window flow control between a transmitter and a receiver consists of an upper bound \ \'.4 [J on the number of data units and permits in transit inside the network