Execution platform modeling for system-level architecture performance analysis
V.D. Zivkovic · Leiden Repository (Leiden University) · 2008
Today's embedded systems are designed for more complex and more computationally-intensive applications than they were a decade ago. Most if not every embedded system designed today is a sort of parallel computing system - only called differently - a platform. A platform is essentially a heterogeneous system consisting of communicating processing units of different types and mostly distributed memory units. A platform can be anything: from multiprocessors comprising task-dedicated processors and a dedicated communication network, to a (semi-)programmable multiprocessor that can run parallel processes by means of both interleaving and overlapping. However, the specification, exploration and design of application multiprocessor system platforms from user requirements is still a painstaking process that takes too long and is too costly. Our answer to the above mentioned issues is the Archer approach. It embodies: Application representations (Symbolic Programs - SP), a platform-based library of the architecture components and their configurations (all-in-hardware, all-in-software, hybrid multiprocessor, with dedicated network, hared-bus, highway, burst-bus, or hybrid network), and a mapping methodology (managing the aforementioned representations while transforming application SPs to Archer architecture components), that we have been developing and experimenting with.