16 Introduction and Layered Network Architecture Chap.1 link capacities in the future (the reason for this is discussed later).Thus circuit switch- ing is a feasible possibility (although not necessarily the best possibility)for networks of the future.Part of the issue here is that as link speeds increase,node processing speed must also increase,putting a premium on simple processing within the subnet. It is not yet clear whether circuit switching or store-and-forward allows simpler subnet processing at high link speeds,but store-and-forward techniques are currently receiving more attention. In the store-and-forward approach to subnet design,each session is initiated without necessarily making any reserved allocation of transmission rate for the session.Similarly, there is no conventional multiplexing of the communication links.Rather,one packet or message at a time is transmitted on a communication link,using the full transmission rate of the link.The link is shared between the different sessions using that link,but the sharing is done on an as needed basis (i.e.,demand basis)rather than a fixed allocation basis.Thus,when a packet or message arrives at a switching node on its path to the destination site,it waits in a queue for its turn to be transmitted on the next link in its path. Store-and-forward switching has the advantage over circuit switching that each communication link is fully utilized whenever it has any traffic to send.In Chapter 3, when queueing is studied,it will be shown that using communication links on a demand basis often markedly decreases the delay in the network relative to the circuit switching approach.Store-and-forward switching,however,has the disadvantage that the queueing delays in the nodes are hard to control.The packets queued at a node come from inputs at many different sites,and thus there is a need for control mechanisms to slow down those inputs when the queueing delay is excessive,or even worse,when the buffering capacity at the node is about to be exceeded.There is a feedback delay associated with any such control mechanism.First,the overloaded node must somehow send the offending inputs some control information (through the links of the network)telling them to slow down. Second,a considerable number of packets might already be in the subnet heading for the given node.This is the general topic of flow control and is discussed in Chapter 6.The reader should be aware,however,that this problem is caused by the store-and-forward approach and is largely nonexistent in the circuit switching approach. There is a considerable taxonomy associated with store-and-forward switching. Message switching is store-and-forward switching in which messages are sent as unit entities rather than being segmented into packets.If message switching were to be used,there would have to be a maximum message size,which essentially would mean that the user would have to packetize messages rather than having packetization done elsewhere.Packet switching is store-and-forward switching in which messages are broken into packets,and from the discussion above,we see that store-and-forward switching and packet switching are essentially synonymous.Virtual circuit routing is store-and-forward switching in which a particular path is set up when a session is initiated and maintained during the life of the session.This is like circuit switching in the sense of using a fixed path,but it is virtual in the sense that the capacity of each link is shared by the sessions using that link on a demand basis rather than by fixed allocations.Dynamic routing (or datagram routing)is store-and-forward switching in which each packet finds its own path16 Introduction and Layered Network Architecture Chap. 1 link capacities in the future (the reason for this is discussed later). Thus circuit switch￾ing is a feasible possibility (although not necessarily the best possibility) for networks of the future. Part of the issue here is that as link speeds increase, node processing speed must also increase, putting a premium on simple processing within the subnet. It is not yet clear whether circuit switching or store-and-forward allows simpler subnet processing at high link speeds, but store-and-forward techniques are currently receiving more attention. In the store-and-forward approach to subnet design, each session is initiated without necessarily making any reserved allocation of transmission rate for the session. Similarly, there is no conventional multiplexing of the communication links. Rather, one packet or message at a time is transmitted on a communication link, using the full transmission rate of the link. The link is shared between the different sessions using that link, but the sharing is done on an as needed basis (i.e., demand basis) rather than a fixed allocation basis. Thus, when a packet or message arrives at a switching node on its path to the destination site, it waits in a queue for its tum to be transmitted on the next link in its path. Store-and-forward switching has the advantage over circuit switching that each communication link is fully utilized whenever it has any traffic to send. In Chapter 3, when queueing is studied, it will be shown that using communication links on a demand basis often markedly decreases the delay in the network relative to the circuit switching approach. Store-and-forward switching, however, has the disadvantage that the queueing delays in the nodes are hard to control. The packets queued at a node come from inputs at many different sites, and thus there is a need for control mechanisms to slow down those inputs when the queueing delay is excessive, or even worse, when the buffering capacity at the node is about to be exceeded. There is a feedback delay associated with any such control mechanism. First, the overloaded node must somehow send the offending inputs some control information (through the links of the network) telling them to slow down. Second, a considerable number of packets might already be in the subnet heading for the given node. This is the general topic of flow control and is discussed in Chapter 6. The reader should be aware, however, that this problem is caused by the store-and-forward approach and is largely nonexistent in the circuit switching approach. There is a considerable taxonomy associated with store-and-forward switching. Message switching is store-and-forward switching in which messages are sent as unit entities rather than being segmented into packets. If message switching were to be used, there would have to be a maximum message size, which essentially would mean that the user would have to packetize messages rather than having packetization done elsewhere. Packet switching is store-and-forward switching in which messages are broken into packets, and from the discussion above, we see that store-and-forward switching and packet switching are essentially synonymous. Virtual circuit routing is store-and-forward switching in which a particular path is set up when a session is initiated and maintained during the life of the session. This is like circuit switching in the sense of using a fixed path, but it is virtual in the sense that the capacity of each link is shared by the sessions using that link on a demand basis rather than by fixed allocations. Dynamic routing (or datagram routing) is store-and-forward switching in which each packet finds its own path