Timeliness
We refer to the operation of an action, because actions that are computational entities which execute in computers are a special case. For example, radars, weapons, robots, factory machines, and many other physical devices perform real-time actions that are not computational entities, and hence operate rather than execute.
Speaking casually, most people implicitly have an informal mental model that considers information or an event as being “real-time”
- if, or to the extent that, it is manifest to them with a delay after its change (i.e., latency) which can be related to its perceived currency
- i.e., manifest in a time frame during which the information or event has acceptably satisfactory usefulness to them.
That mental model is meaningful, but reasoning about it requires additional formalization.
More technically, a first principle based mental model for a real-time action is that: the manifestation is completion of the action’s operation; latency from initiation of the action is delineated with a deadline; and currency (acceptability of satisfaction) is determined by the action’s completion time with respect to (e.g., either early before, or tardy after) its deadline.
That definition of a deadline is the (simplified) standard scheduling theory one. The real-time computing community thinks in terms of an incomplete special case (i.e., only “meet” vs. “miss) as described in Chapter 2.
Even the scheduling theory definition of deadline is limited by its linearity of lateness. In general, acceptably satisfactory usefulness of an action can be non-linear with respect to its completion time. That can be represented with a richer mental model which expresses the usefulness of an action’s completion time as an application-specific function—i.e., a time/utility function (TUF) [Jensen 77, Jensen 85]. In the TUF model, a deadline function has a downward step from 1 to 0 at the deadline time. The TUF model is described in Chapter 2.
A different mental model at the other extreme from latency with respect to a deadline is that there is no deadline, and that the shorter the latency, the greater the satisfaction. That is a simplified description of mental models in which latency is regarded for actions (and systems) based on patterns such as publish/subscribe—e.g., the OMG DDS standard [ ] and its many predecessors (e.g., [ ]).
The magnitudes of the latency time frames in both of these mental models, whether microseconds or megaseconds, are application- and situation-specific, and are not properly part of the definition of real-time—contrary to wide-spread misconception in the real-time computing field (e.g. [ ]).
It is easy to think of various circumstances in either real-time computing or real life that have widely different action time frames and entirely different relationships between action manifestation latencies and action values (similarly at the system level). For example, in military defense, the notional alert “incoming!” implies quite different time frames and latency-based action (and system) satisfaction values when referring to a sea-skimming anti-ship missile vs. a land-based and land-targeted ballistic missile. In everyday life, a telephone call about your child at school implies quite different time frames and latency-based action (and system) satisfaction values when it refers to your child being injured vs. your child needing to bring lunch tomorrow.
The two real-time action mental models as expressed here informally but clearly capture the first principle latency-based satisfaction property of real-time actions (and systems) as timeliness.
But real-time is not just about timeliness of action completion, it has a second equally fundamental distinguishing property—predictability of that timeliness.