Modeling and performance evaluation of UML-based platform independent models for MPSoC embedded software

Abstract

The MPSoC has become a solution for designing embedded systems especially dedicated to applications that require intensive computations. As it gives more importance to software than before, the complexity of embedded software is growing rapidly, which brings embedded software design. Designing embedded software conforming various hardware platform constraints is difficult. To deal with the problem, the concept of Model-Driven Architecture(MDA) has been introduced in embedded software design. As the initial step of the MDA approach, the essential features of the system is captured by the Platform Independent Model(PIM). This thesis focuses on the PIM because it dominates the whole MDA-based software development process as the starting point. For the description of PIM, the Unified Modeling Language(UML) is widely used. To enable early design decisions in the development of embedded software, a method to check not only the functional requirements but also non-functional ones for the UML-based PIM is needed. In this thesis, we suggest an approach to modeling and performance evaluation of UML-based PIM for MPSoC embedded software. We first suggest an MDA-based design flow for MPSoC embedded software and explain which part of the flow we are focusing on. Then we present how we model PIM from use case model. And we describe performance evaluation of PIM. The evaluation is performed by simulation. For the simulation, we first perform control flow analysis on the PIM and identify tasks. Then we construct Platform Model(PM) to model the target platform where the software is supposed to be deployed, and allocate tasks to hardware resources in PM. Regarding the allocation, we suggest a heuristic algorithm to automatically allocate tasks to hardware resources. The allocation by our algorithm can be used as a guide to the designer. After the allocation, we simulate the PIM and estimate performance. In the simulation, we use abstract information of PM. With the informatio...