Long distance switches may use a slower, more efficient switch-allocation algorithm than local central offices, because they have near 100% utilization of their input and output channels. Central offices have more than 90% of their channel capacity unused.
While traditionally, telephone switches connected physical circuits (e.g., wire pairs), modern telephone switches use a combination of space- and time-division switching. In other words, each voice channel is represented by a time slot (say 1 or 2) on a physical wire pair (A or B). In order to connect two voice channels (say A1 and B2) together, the telephone switch interchanges the information between A1 and B2. It switches both the time slot and physical connection. To do this, it exchanges data between the time slots and connections 8000 times per second, under control of digital logic that cycles through electronic lists of the current connections. Using both types of switching makes a modern switch far smaller than either a space or time switch could be by itself.
The structure of a switch is an odd number of layers of smaller, simpler subswitches. Each layer is interconnected by a web of wires that goes from each subswitch, to a set of the next layer of subswitches. In most designs, a physical (space) switching layer alternates with a time switching layer. The layers are symmetric, because in a telephone system callers can also be callees.
A time-division subswitch reads a complete cycle of time slots into a memory, and then writes it out in a different order, also under control of a cyclic computer memory. This causes some delay in the signal.
A space-division subswitch switches electrical paths, often using some variant of a nonblocking minimal spanning switch, or a crossover switch.
Switch control algorithms
Fully-connected mesh network
One way is to have enough switching fabric to assure that the pairwise allocation will always succeed by building a fully-connected mesh network. This is the method usually used in central office switches, which have low utilization of their resources.
Clos's nonblocking switch algorithm
The scarce resources in a telephone switch are the connections between layers of subswitches. The control logic has to allocate these connections, and most switches do so in a way that is fault tolerant. See nonblocking minimal spanning switch for a discussion of the Charles Clos algorithm, used in many telephone switches, and a very important algorithm to the telephone industry.
Fault tolerance
Composite switches are inherently fault-tolerant. If a subswitch fails, the controlling computer can sense it during a periodic test. The computer marks all the connections to the subswitch as "in use". This prevents new calls, and does not interrupt old calls that remain working. As calls in progress end, the subswitch becomes unused, and new calls avoid the subswitch because it's already "in use." Some time later, a technician can replace the circuit board. When the next test succeeds, the connections to the repaired subsystem are marked "not in use," and the switch returns to full operation.
To prevent frustration with unsensed failures, all the connections between layers in the switch are allocated using first-in-first-out lists. As a result, if a connection is faulty or noisy and the customer hangs up and redials, they will get a different set of connections and subswitches. A last-in-first-out allocation of connections might cause a continuing string of very frustrating failures.
No comments:
Post a Comment