On the technical side the ultimate goal for the project is the reduction of the remaining uncertainty of real-time system behaviour and the reduction of the penalties to pay for this uncertainty.
The metric in which predictability is measured is the resulting degree of overestimation. Hence, the progress resulting from the project will be measured in the development of this degree.
Since overestimation can be understood as Uncertainty × Penalties, our goal consists of two subgoals:
Each technique proposed in the project will be evaluated by measuring the reduction in the overestimation with respect to the case in which that technique is not used. Efficiency of new designs will be measured as usual by simulation and profiling.
Technological trends, e.g. the increasing processor-memory speed gap, will work against the goals of the project as has been the case during the past 10 years. Keeping the degree of overestimation at the level that is currently achievable by state-of-the-art analysis tools and considered highly satisfactory by industrial customers, is the most modest promise. We plan to do better on disciplined designs.
It should be noted that in some cases, the progress in the state of the art will consist in making systems analysable that are currently not, e.g. systems employing virtual memory.
Another goal PREDATOR is aiming at is the reduction in the effort necessary to produce performance-analysis tools. An architecture or a system-design discipline with improved predictability properties will need less effort in the development of performance-analysis tools and thus lead to faster time to market for tools and designs produced by using them.
Yet another goal will be to establish a new discipline “Design for predictability in the embedded systems domain”, creating and spreading the awareness for the problems associated with predictability. This goal will be achieved by publications in the relevant conferences and journals and by organising workshops and tutorials.