next up previous contents index
Next: Definition Up: Random Variables Previous: Random Variables   Contents   Index


Overview

In the Palladio Component Model, we use random variables in expressions which specify parametric dependencies. The rationale behind this is that many aspects of larger software systems, especially in the business information systems area, can not be modelled having complete information (see for a classification of the information types for example [15]). Uncertainties can be found in many aspects of the system model. Two main sources stem from the behaviour of users and time spans of method executions (because we do not consider real-time environments). User behaviour can only be specified in a stochastical manner. How long users think (aka think time) between requests to the system, what parameter values they use in their requests can often only be characterized using probabilities. The second source comes from the facts that the PCM is designed to support predictions on an architectural level. On such a level, real-time constraints are usually not available. The reasons are also manifold. First, the information is simply not available at early design stages. Second, environmental features as garbage collectors, middle-ware services, etc. make it hard to predict timing time consumptions with certainty.

In the following, the use of random variables in parametric dependencies is introduced. These types of specifications are used in several places in the PCM. They are used especially in the ResourceDemandingSEFF and the ResourcePackage to describe resource consumptions. A library in the implementation of the PCM supports the use of random variables in different types. See 4.1.4 for technical details on this library.


next up previous contents index
Next: Definition Up: Random Variables Previous: Random Variables   Contents   Index
Snowball 2007-03-16