Abstract
An asymmetric multi-processor system is usually composed of a general pur- pose RISC processor and one or more con‾gurable computing unit such as DSP cores. While asymmetric multi-processor system platform is getting much an much more popular on most modern embedded devices, many researches are devoted to constructing a °exible scheduling framework on such a platform. By parallel exe- cuting tasks between di®erent computing units, we may be able to use two or more lower speed processors than the traditional one high speed uni-processor system, if and only if we schedule the tasks right on the suitable kind of processor. Every kind of processors has its own power and speed relationship. Every task due to di®erent implement algorithm and running on di®erent hardware implementation also encounters di®erent execution clock cycle counts and spent di®erent quantity of energy. The more clock cycle counts one processor should execute in one period, the more speed and energy consumption a processor will have. How to equally and e±ciently schedule tasks between heterogeneous processors to achieve a su±cient performance requirement while making the energy consumption of the system min- imum is our goal. This paper provide a task decision method that can evaluate the energy performance index of one task on each processor under a frame based real- time system. Besides we provide another load-balance algorithm which is needed when the scheduling result is under an un-balanced load situation that might waste the computing power of another processor.