network delay Broadly speaking, the time needed for a signal to traverse a network. The extent of the delay caused by the network may be effectively constant, but in most cases it is variable. If the variable delay can be guaranteed not to exceed some predetermined value, then the network has a bounded delay. In other cases the delay is not bounded but can grow without limit, although with a decreasing probability of a longer delay.

In a circuit switching network the only significant delay arises from the finite speed with which the signals propagate along the transmission medium. For electrical signals on a conducting wire, electromagnetic waves in free space, or light signals in an optical fiber, this speed is of the same order as the speed of light, 300 000 km per second, and the network delay in a circuit-switched system is of the order of 3–5 microseconds per km. There may also be delays arising from the finite time needed by amplifiers or repeaters to pass a signal from their input to output; these delays are again in the order of microseconds. The delay in a signal crossing the Atlantic (5000 km) is of the order of 25 milliseconds, and for a signal routed via a geostationary satellite (total round trip of 80 000 km) it is of the order of 400 milliseconds.

In a packet switching network the situation is more complex. The time needed to traverse the network is normally measured as the period between the sender indicating that the transmission is to start, and the delivery of the last bit of the packet to the destination. This time is the sum of times needed to traverse each sector of the network, and contains a number of different contributions. If the data source is not capable of generating a network packet directly, it will need to be connected to a PAD that will assemble the data into packets; devices that can generate their own packets will not require a PAD. It may well be that although a PAD is not present as a separate component, it is still there conceptually, where for example a (human) user is using a PC or workstation to connect to a remote system. Once packets have been generated, each packet will move between successive pairs of network nodes until it reaches the destination. Each sector has contributions from

(a) the transit time along the medium connecting the two nodes;

(b) the time needed to disassemble the outgoing packet into its component bits at the transmitting node (necessarily identical with the time needed to reassemble the incoming packet at the receiving node);

(c) the time needed by the switching process within the receiving node to determine the route the outgoing packet is to follow, and carry out the switching.

The first of these is essentially similar to the transit time in a circuit-switched network, and has a similar value of say 3–5 microseconds per km. The second is essentially equal to the packet size multiplied by the inter-bit time on the transmission line. The third is a function of the organization of the switching nodes, of their processing speeds, and of the extent to which the switching must be delayed until the information needed to allow switching to start is determined by the internal structure of the packet. The use of cell relay systems, in which switching can start before the entire packet has been received, allows this time to be reduced.

A heavily loaded system will have queues (lines) of packets in each node, leading to a further complication. The queue may either be of outgoing packets awaiting the attention of the switching process to determine on which onward connection they should be transmitted, or of packets that have been rerouted to an onward connection that is already active, requiring the packet to wait until those ahead of it have been transmitted.

In networks using only terrestrial links, the total network delay is typically dominated by packet assembly times, (b) above. This is not the case where satellite links are used, especially where geostationary satellites are involved.