Resource allocation framework for distributed real-time end-to-end tasks

Traditional resource allocation algorithms such as Q-RAM (Rajkumar et al., 1997) assume that importance (or weight) or utility values for distributed real-time tasks is a totally ordered set to measure the rewards for completing every task. Hence, resource allocation problem can be viewed as the problem of maximizing total utility values. However, in several real-time applications such as multi-function phased array radar (MFPAR) systems, totally ordered importance are not available. We develop a two-level resource allocation framework. The framework allows the schedulers for subsystems or processors in distributed realtime systems to autonomously schedule local sub-tasks and the system performance is enhanced without heavy global optimization overhead. In addition, the framework can trade the run-time overhead including time and memory space with the optimality of resource allocation. We evaluate our framework by extensive simulations for MFPAR systems. The experimental results show that the developed framework outperforms the traditional priority-based approach.