Everything within a computer runs in lock-step, those steps being dictated by the ticking of an extremely fine-resolution internal clock -- these ticks flow around the computer and provide a synchronization point for all of the units, so that they all "know" exactly what time it is, and can use that knowledge to control their own activities. Obviously, the faster the clock ticks, the faster work can be accomplished, assuming that those doing the work are up to meeting the faster beat. Still, that's what's meant by increasing the clock: that technology has enabled the working units in the processor to perform meaningful computation at a higher rate of speed ... there's no trick to generating a fast signal, of course, the trick is in making sure everyone else is going to be able to take advantage of it.

Commodity processors have clocks pushing 100MHz (mega-hertz), or capable of generating a pulse every 10 nano-seconds (10-9 seconds); state-of-the-art processors are at or above 250MHz (4 ns) and higher ... some are approaching 1000MHz (1 ns), and we can expect to see sub-ns ( <1ns.) probably before the Millenium.

The faster the clock ticks, the faster instructions get issued and are ready for execution, and the more instruction units within the processor, the more instructions are ready for execution simultaneously. A high percentage of these operations, at any point in time, will be loads or stores, operations requiring access to memory, so, in order to keep things moving as quickly as possible, the ability of the processor to obtain access to any particular location in memory must approach the limit set by assuming that every operation is a memory request -- thus the scaling referred to above: the number of things issuing instructions times the number of times instructions can be issued.