Real-Time Systems. Design Principles for Distributed Embedded Applications. Herman Kopetz. Second Edition (811374), страница 18
Текст из файла (страница 18)
Therequirement to build artifacts, the properties of which can be analyzed by simplemodels, should thus be an explicit design driver. In many areas of computer sciencethis principle of building artifacts that can be modeled by simple models is violated.2.4 Emergence43For example, the temporal behavior of a modern pipelined microprocessor withmultiple caches cannot be captured in a simple model.The major challenge of design is the building of a software/hardware artifact (anembedded computer system) that provides the intended behavior (i.e. the service)under given constraints and where relevant properties of this artifact (e.g., thebehavior) can be modeled at different levels of abstraction by models of adequatesimplicity.As stated before, there are many different purposes that give rise to a hierarchyof models of an artifact.
Examples are: behavior, reliability, man–machine interaction, energy consumption, physical dimension, cost of manufacturing, or cost ofmaintenance, to name a few. Out of these, the most important one is the model ofbehavior. In the context of real-time systems, behavior specifies the output actionsof a computer system as a consequence of the inputs, the state and the progressionof real-time. Output actions and input can be captured in the concepts of inputmessages and output messages.
In Chap. 4 of this book we present a cross-domainmodel for the behavior of a real-time computer system using these concepts.2.4EmergenceWe speak of emergence when the interactions of subsystems give rise to uniqueglobal properties at the system level that are not present at the level of thesubsystems [Mor07]. Non-linear behavior of the subsystems, feedback and feedforward mechanisms, and time delays are of relevance for the appearance ofemergent properties. Up to now, the phenomenon of emergence is not fullyunderstood and a topic of intense study.2.4.1IrreducibilityEmergent properties are irreducible, holistic, and novel – they disappear when thesystem is partitioned into its subsystem.
Emergent properties can appear unexpectedly or they are planned. In many situations, the first appearance of the emergentproperties is unforeseen and unpredictable. Often a fundamental revision ofstate-of-the-art models is required to get a better understanding of the conditionsthat lead to the intended emergence. In some cases, the emergent properties can becaptured in a new conceptualization (model) at a higher level of abstraction resulting in an abrupt simplification of the scenario.Example: The emergent properties of a diamond, such as brilliance and hardness, whichare caused by the coherent alignment of the Carbon-atoms, are substantially different fromthe properties of graphite (which consists of the same atoms).
We can consider the diamondwith its characteristic properties a new concept, a new unit of thought, and forget about itscomposition and internal structure. Simplicity comes out as a result of the intricate interactions among the elements that help to generate a new whole with its new emergent properties.442 Simplicity2.4.2Prior and Derived PropertiesWhen dealing with emergence, it is helpful to distinguish between the priorproperties of the components and the new derived properties that come about bythe interactions of the components.Example: The high reliability of the services of a fault-tolerant system (derived property)that is the result of the interactions of many unreliable components (prior property) is anemergent property.In many cases the prior properties and the derived properties can be of a completelydifferent kind.
It often happens that the derived properties open a completely newdomain of science and engineering. This new domain requires the formation ofnovel concepts that capture essential properties of this new domain.Example: The property of being able to fly which comes about by the proper interaction ofthe subsystems of an airplane, such as the wings, the fuselage, the engines and the controls,is only present in the airplane as a whole but not in any of the isolated subsystems. Beingable to fly has opened the domain of the air transportation industry with its own rules andregulations. For example, the subject of air traffic control is far removed from the priorproperties of the components that make up an airplane.Prior properties and derived properties are relative to the viewpoint of the observer.When climbing up the abstraction ladder, the derived properties at one level ofabstraction become the prior properties at the next higher level of abstraction and soon, since a new form of emergence can appear at higher levels.Example: In the evolution of the universe two very significant stages of emergence are theappearance of life and at a further stage the appearance of consciousness that forms the basisfor the development of human culture.
The realm of human culture has developed its ownsystem of concepts in the arts, sciences etc., that are far removed from the biological priorproperties that are characterizing the human brain.Emergent behavior cannot be predicted analytically, but must be detected in anoperating system. Thus control elements must incorporate hooks for monitoringsystem performance in real time [Par97, p. 7]. The multicast message concept,discussed in Sect. 2.2.3 provides the basis for the nonintrusive observation ofsystem behavior.2.4.3Complex SystemsWe classify a system as complex if we are not in the position to develop a set of modelsof adequate simplicity – commensurate to the rational capabilities of the humanmind – to explain the structure and behavior of the system.
In addition to life andconsciousness, examples for complex systems are the earth’s climate and weather, theglobal economy, living organisms, and many large computer systems, to name a few.We hold the opinion that a fundamental understanding of a complex system canonly be achieved by a proper conceptualization and not by the execution of2.5 How Can We Achieve Simplicity?45elaborate computer simulations.
This view is also shared by Mesarovic et al.[Mes04, p.19] when he speaks about biology:We further argue that for a deeper understanding in systems biology investigations shouldgo beyond building numerical mathematical or computer models – important as they are. . .. Such a categorical perspective led us to propose that the core of understanding insystems biology depends on the search for organizing principles rather than solely onconstruction of predictive descriptions (i.e. models) that exactly outline the evolution ofsystems in space and time. The search for organizing principles requires an identification/discovery of new concepts and hypotheses.Maybe, sometimes in the future, we will form appropriate concepts that will lead toan abrupt simplification of some of today’s complex systems.
If this happens, thesystem will not be classified as complex any more.Whereas system biology deals with a natural system, a large computer system isan artifact developed by humans. When designing such an artifact, we should takeconsideration of the limited rational problem solving capability of humans in orderthat we can describe the behavior of the artifact by models of adequate simplicity.These models should guide the design process, such that a structure clash betweenthe model and the artifact is avoided.Example: Let us look at the technical example of designing the on-chip communicationinfrastructure for the communication among IP-cores on a system-on-chip. There arebasically two technical alternatives, the provision of a shared memory that can be accessedby all IP-cores or the provision of local memory to each one of the IP-cores and the designof a message-passing subsystem that enables the exchange of messages among IP-cores[Pol07,Lev08].
The message-passing subsystem isolates and makes explicit the globalcommunication among subsystems and thus supports the introduction of a new level inthe hierarchy where a distinction is made between the intra-IP core interactions and theinter-IP core interactions. The common memory intermixes global intra-IP-core and localinter-IP-core interactions and makes it very difficult to separate global and local concerns,leading to a more complex system model.2.5How Can We Achieve Simplicity?Cognitive scientists have studied how students learn and understand different tasks[Fel04]. They have identified a set of task characteristics that require a disproportionalmental effort for understanding the task.
Table 2.2 compares the characteristics ofsimple tasks versus difficult tasks. We thus need to design a generic model forexpressing the behavior of an embedded system that avoids the characteristics ofdifficult tasks. It should be possible to apply the model recursively, such that largesystems can be modeled at different levels of abstraction using the same modelingmechanisms.The model of a real-time system, presented in Chap. 4, tries to reach this goal.Simplicity is achieved by adhering to the following seven design principles:1. Principle of Abstraction. The introduction of a component (a hardware/software unit) as a basic structural and computational unit makes it possible to use462 SimplicityTable 2.2 Characteristics of simple versus difficult tasks (Adapted from [Fel04, p.
91])Characteristics of a simple taskCharacteristics of a difficult taskStatic: The properties of the task do not change Dynamic: The properties of the task are timeover time.dependant.Discrete: The variables that characterize theContinuous: The domain of the variables istask can only take values from discrete sets.continuous.Separable: Different subtasks are nearlyNon-separable: Different subtasks are highlyindependent. There is only a weakinteractive.
![](https://s.studizba.com/z.php?f=/templates/si/i/noavatar.png&w=100&h=100&t=1"){kind=avatar}
