In this final section, we briefly discuss ways in which all of the previous elements can be physically interconnected. Over the decades this infrastructure has evolved from the centralized mainframe architecture, with terminals attached via RS.232 or RS.422 (or some proprietary means) through to a wide armory of local and wide area connectivity options.
Today there are many ways to interconnect network elements, each with its own particular strengths and weaknesses. The following list is far from exhaustive:
Low-speed, direct-attached, serial lines (RS.232, RS.422, etc.)
Dial-up serial lines (analog and digital)
Local Area Networks (Ethernet, Token Ring, AppleTalk, ATM)
Metropolitan Area Networks (FDDI, SMDS, ATM, SONET/SDH)
Wide Area Networks (Frame Relay, ATM, ISDN, X.25, leased lines, satellite links)
Wireless LANs and WANs (802.11, Bluetooth, GSM, GPRS, UMTS)
There are marked differences among many of these technologies; some of the key differentiators are as follows:
Packet, cell, or circuit switching
Wired or wireless
Quality of Service (QoS)
To the uninitiated the choice can be bewildering, and even more perplexing is the fact that many countries have different standards, different charging models, and different service availabilities. In Figure 1.15, we see that a variety of packet-switched (ps), circuit-switched (cs), and point-to-point (pp) technologies are available, each applicable for the LAN, MAN, WAN, or all environments. Although most of these technologies have a range of bandwidth offerings, the choice has continued to expand vertically and horizontally over the past 40 years, beginning with simple leased circuits back in the 1960s.
Figure 1.15: Sample media and bandwidth choices available today.
One of the key differentiators between data network connectivity is whether the technology is circuit or packet switched.
Circuit switching has its history firmly rooted in voice networks. In the telephone network Pulse Code Modulation (PCM) is used to transmit digital information by sampling voice at 8,000 times per second (corresponding to 125 µs per sample), where each sample is encoded as an 8-bit number (note that this is where the magical 64-Kbps circuit multiplier originates: 8 X 8,000 = 64,000 bps). Unlike packet switching, this digital information does not require a protocol header and addressing information to be routed. In order to transfer data between two locations a call is made to the destination node. This call reserves a discrete set of physical resources (circuits or channels) across the network; only when these resources have been allocated can data be transferred. Routing is therefore implicit in the physical path. Although the circuit is dedicated, on large trunk connections traffic can be interleaved with other users' traffic through strict timing control (called multiplexing). Once the data transfer is complete, the call may be disconnected. Note that this model supports permanent calls if required, and either packetized data or digitized voice can be carried over circuit-switched networks. Examples of circuit-switching technologies include voice calls, analog dial-up circuits, and ISDN.
Packet switching covers packet, frame, or cell switching (the main difference being that cells have fixed length, and both packets and frames are variable). In packet-switched networks each packet is an autonomous unit (called a datagram), with source and destination address information placed in a header preceding each packet. When data are transferred between two nodes, each packet traverses the network following logical paths created by intermediate nodes (such as routers or switches). Packets are normally interleaved with traffic from other sources, although there is no strict timing required, and packets may travel autonomously down different routes to reach their destinations. Examples of packet-switching technologies include X.25, Frame Relay, ATM, and SMDS.
There are fundamental differences in the way the circuit- and packet-switched networks behave under load. In essence, with circuit switching, network resources are either utilized or they are not. While a circuit and associated switch ports are in use, the subscriber has exclusive access to that resource; the circuit is effectively dedicated. In heavy subscription periods it is, therefore, possible that all resources will be allocated (referred to as a period of oversubscription). In this event new subscribers are blocked from making new calls until circuits are freed up, even though in practice many circuits could be underutilized (i.e., any spare bandwidth is locked up). A circuit-switched network, therefore, needs to have sufficient circuit and switching equipment capacity to cope with heavy use periods in order to keep customers satisfied. This can lead to substantial resource waste and cost inefficiencies during average/low use periods, leading to a practice referred to as overprovisioning.
By contrast, with a packet-switched network, traffic is generally competing for shared circuit resources. As more users subscribe to the network, circuit utilization increases and performance gradually degrades. As resources approach saturation, packets are simply discarded (either because they cannot be serviced or because they are aged out of queues); new users are not explicitly blocked. Overall the network can be better optimized through traffic-engineering techniques, enabling service providers to provision the network more cost effectively. To enable different service classes to be supported, soft policy control can be imposed over the whole network (in which case best-effort packets could be discarded prior to saturation to alleviate congestion problems).
Note that these two switching techniques are applicable to both wired and wireless network solutions. We will discuss these issues in more depth later in the book.
Providing connectivity, whether packet or circuit switched, is only half of the problem. To satisfy traffic requirements we also need to provide an appropriate level of bandwidth. The technologies we have introduced so far operate at different speeds, and the methods used to access the physical media, the quality of the media, the distances traveled, and the protocols used to transfer data all have a part to play in determining the effective rate at which data can be transferred. Figure 1.15 illustrates a selection of the key local, metropolitan, and wide area networking technologies available today and their respective bandwidth ranges. The unique operational characteristics of these technologies often mean that different design techniques must be used in each case. The various design implications are discussed in detail later in the book.