Switched Networks: A Refresher

Switched networks always involve switches of some sort that are interconnected by circuits. At the simplest level, a switch serves to interconnect circuits through a switching matrix. The circuits terminate in ports, which are physical points of interface, and which typically are directional (i.e., either incoming or outgoing). The switch receives incoming traffic on an incoming port, analyzes the destination address (i.e., the address for which the call is intended), perhaps also analyzes the originating address, consults an internal routing table, determines the appropriate outgoing port associated with an outgoing circuit, sets up a path through the switching matrix between the two ports, and sends the data on its way. The path through the switching matrix is a logical path that comprises one or more internal physical circuits.

All contemporary switches have certain things in common. They all are Electronic Common Control (ECC), are digital in nature and are specialized computer systems. Like any other computer system, they consist of cabinets, shelves, printed circuit boards, port interfaces, power supplies and so on. The common control element of a switch is in the form of a set of programmed logic stored in memory and associated with multiple microprocessors that form a Centralized Processing Unit (CPU) that actually may be distributed (i.e., decentralized). Taken together, the common control element controls the operation of the system and its various components. The switching matrix comprises an intricately interconnected set of internal circuitry that connects incoming ports to outgoing ports. The port interfaces serve to establish physical connectivity between the switch and the external circuits, which may be either metallic or optical in nature.

In the Wide Area Network (WAN) domain, edge switches are positioned at the edge of the network, serving to provide access to the network from the customer premises over local loops. In contemporary terms, these edge switches commonly are known as POPs (Points of Presence). In the core, or backbone, the WAN core switches, also known as backbone switches or tandem switches, are positioned. Core switches serve no end users directly. Rather, they serve to interconnect lesser edge or core switches and interconnect with their peers over high-capacity circuits.

In the Local Area Network (LAN) domain, workgroup switches are equivalent to edge switches in the WAN, as they serve end users directly. The workgroup switches may be interconnected directly, or through a backbone switch.

Circuit Switches

Circuit switches came first, having been developed in 1878 for use in the Public Switched Telephone Network (PSTN). Generally speaking, circuit switches are optimized for uncompressed, realtime voice communications, which is stream-oriented and which is very demanding in terms of its flow and pace. In consideration of the demands of the primary application, circuit-switches are truly connection-oriented. That is to say that a predefined path is established through the network, from the incoming port on the originating edge switch to the outgoing port on the terminating edge switch, and through each and every intermediate core switch and across every circuit involved. While each switch acts independently in establishing its portion of the overall path, they usually act in concert under the command of a centralized signaling and control system such as Signaling System 7 (SS7).

Once the circuit switches have established a connection, that connection is maintained on a temporary, continuous, and exclusive basis. The connection is temporary, as the path and its constituent portions are maintained only as long as both the transmitting and receiving terminals are engaged (i.e., off-hook, in voice terms). The connection is continuous, as it is continuously available in support of transmission in both directions, whether or not any real information is being transmitted.

In other words, silence bytes are considered just as important as sound bytes. The connection, or path, is exclusive, as it is shared with no other transmissions. That is not to say that other conversations are not traveling the same physical path, but it is to say that no other conversations share the same logical channels over the same physical circuits, or paths. In other words, networks resources are committed to your conversation (of sound bytes and silence bytes), whether you are talking, listening, or sleeping on the phone. The connection is committed, as are all of the involved resources, from switches to circuits and channels -- and you pay for the privilege, on the basis of so much per minute or fraction thereof.

Circuit switches offer outstanding performance. In fact, they offer virtually perfect performance. But, they are highly inefficient for anything other than realtime, uncompressed voice or video.

Packet Switches

Packet switches made their debut in 1971, for use in the ARPANet (Advanced Research Project Agency Network), which later spawned the commercial Internet. This packet technology was developed specifically in support of interactive data communications between asynchronous host computers and terminals. The application was what we then called time-sharing, which involved dumb remote terminals sharing access to files and applications residing on very expensive centralized mainframe host computers. (Note: Those huge mainframes had a fraction of the horsepower of your laptop computer.)

Packet switches act to switch data in discrete units known as packets, or datagrams, each of which is of a fixed minimum and maximum size. The specific size of the packet is established for each network, and can vary from network to network in an interconnected network (i.e., internet) environment, although there always are fixed minimums and maximums.

Packets are individually addressed, which allow the packet switches to deal with each independently to forward it along the most appropriate and available path. As each packet works its way through the network independently, and as each can take a different path through the various switches and across the various circuits that interconnect them, the network is highly shared-much more so than is a circuit-switched network.

This extreme sharing of resources translates into greater efficiency and lower associated costs. The tradeoff is that performance can suffer as packets experience varying levels of latency, or delay, as they work their way through the network. That latency can be the result of some combination of factors:

• Route length affects latency, as a longer physical route increases the time it takes the signal to propagate, or transverse, across the network.
• Hop count affects latency as the greater the number of switches involved, the more times a packet must be analyzed and acted upon, all of which takes time.
• Congestion is an unpredictable variable as sometimes the shared network is highly available and the packets can work their way across the network fairly immediately, and sometimes it is heavily used and the packets must wait in queues for various lengths of time.
• Individual packets may require retransmissions to recover from damage, loss, or delay that exceeds certain limits.