Conservative and compositional modeling in the CompSOC platform
Tal Milea · TU/e Research Portal · 2012
In the CompSOC platform applications are conservatively modeled and analyzed to achieve strict realtime guarantees. Time-non-deterministic system-components are abstracted to their worst-case temporal behavior, resulting in a deterministic Synchronous Dataflow (SDF) model. The individual deterministic models are composed to a system model on which efficient analysis can be performed. An enabling property for ensuring the conservativeness of the analysis performed on the composite-model is the monotonicity of the individual components. Some components in CompSOC have been shown to exhibit non-monotonic behavior, hence need to be treated separately. We propose a framework for the abstraction of non-monotonic components into SDF graphs. The abstraction is compositional in the sense that it allows for abstractions of individual components in a composite-system while preserving the conservativeness of the model. We use the framework to abstract non-monotonic components in the CompSOC communication-channel into a monotonic SDF graphs, which fits well within the CompSOC modeling framework. To complement the modeling of the non-monotonic components of the communication-channel, we provide SDF models of all components of the channel. Our models are either new, or refinements the existing models which provide tighter analysis. Also, we develop a modeling technique that enables modeling of behavior that could not be modeled before. Namely, we model the interference of tasks running on the platform and sharing a buffer or a request-queue under static-order scheduling. This modeling technique can also be applied to general SDF modeling, outside the CompSOC framework.