In an instance where multiple users share the same common resource for communication, oversubscription refers to the amount of guaranteed bandwidth a user will get. Oversubscription is practiced on the basis that it is statistically unlikely that all users will be attempting to use this bandwidth simultaneously, causing noticeable issues.
Oversubscription is often expressed in percentages which come from a ratio, such as 100 or 200% subscription. For example, subscribing two 32K bit/sec permanent virtual circuits to a 64K bit/sec port is 100% subscription, and represents no oversubscription; both circuits could run to their max and the port on the carrier's frame-relay access switch would comfortably have the capacity to support them. However, assigning four 32K bit/sec virtual circuits to that same 64K bits/sec port, is 200% subscription (2-to-1 oversubscription) - at greater, although not great, risk for an issue.
Oversubscription planning and execution is as much an art as it is a necessity. It is not economically practical, environmentally reasonable, or technically feasible to provide dedicated access for every service to every customer. A well-engineered oversubscribed service appears to function as a dedicated service to a subscriber. Another way of saying this is that a good engineer designs the oversubscription in a way that makes the service appear to be a dedicated service for that customer. That means no dropped calls on your cell phone, no busy signals when dialing, and no failed downloads on the Internet.