Thermal control is crucial to real-time systems as excessive processor temperature can cause system failure or unacceptable performance degradation due to hardware throttling. Real-time systems face significant challenges in thermal management as they must avoid processor overheating while still delivering desired real-time performance. Furthermore, many real-time systems must handle a broad range of uncertainties in system and environmental conditions. To address these challenges, this paper presents Thermal Control under Utilization Bound (TCUB), a novel thermal control algorithm specifically designed for real-time systems. TCUB employs a nested feedback loop that dynamically controls both processor temperature and CPU utilization through task rate adaptation. Rigorously modeled and designed based on control theory, TCUB can maintain both desired processor temperature and CPU utilization, thereby avoiding processor overheating and maintaining desired soft real-time performance. A salient feature of TCUB lies on its capability to handle a broad range of uncertainties in terms of processor power consumption, task execution times, ambient temperature, and unexpected thermal faults. The robustness of TCUB makes it particularly suitable for real-time embedded systems that must operate in highly unpredictable environments. The advantages of TCUB are demonstrated through extensive simulations under a broad range of system and environmental uncertainties.

Feedback control techniques have recently been applied to a variety of real-time systems. However, a fundamental issue that was left out is guaranteeing system controllability and the feasibility of applying feedback control to such systems. No control algorithms can effectively control a system which itself is uncontrollable or infeasible. In this paper, we use the multiprocessor utilization control problem as a representative example to study the controllability and feasibility of distributed real-time systems. We prove that controllability and feasibility of a system depend crucially on end-to-end task allocations. We then present algorithms for deploying end-to-end tasks to ensure that the system is controllable and utilization control is feasible for the system. Furthermore, we develop runtime algorithms to maintain controllability and feasibility by reallocating tasks dynamically in response to workload variations, such as task terminations and migrations caused by processor failures. We implement our algorithms in a robust real-time middleware system and report empirical results on an experimental test-bed. We also evaluate the performance of our approach in large systems using numerical experiments. Our results demonstrate that the proposed task allocation algorithms improve the robustness of feedback control in distributed real-time systems.

Feedback control techniques have recently been applied to a variety of real-time systems. However, a fundamental issue that was left out is guaranteeing system controllability and the feasibility of applying feedback control to such systems. No control algorithms can effectively control a system which itself is uncontrollable or infeasible. In this paper, we use the multi-processor utilization control problem as a representative example to study the controllability and feasibility of distributed real-time systems. We prove that controllability and feasibility of a system depend crucially on end-to-end task allocations. We then present algorithms for deploying end-to-end tasks to ensure the system is controllable and utilization control is feasible for the system. Furthermore, we develop runtime algorithms to maintain controllability and feasibility by reallocating tasks dynamically in response to workload variations such as task terminations and migrations caused by processor failures. We implement our algorithms in a robust real-time middleware and report empirical results on an experimental test-bed. Our results demonstrate that the proposed task allocation algorithms improve the robustness of feedback control in distributed real-time systems.

Standards-based quality of service (QoS)-enabled component middleware is increasingly being used as a platform for developing distributed real-time embedded (DRE) systems that execute in open environments where operational conditions, input workload, and resource availability cannot be characterized accurately a priori. Although QoS-enabled component middleware offers many desirable features, until recently it lacked the ability to efficiently allocate resources and configure platform-specific QoS settings based on utilization of system resources and application QoS. Moreover, it has also lacked the ability to monitor and enforce application QoS requirements. This paper presents two contributions to research on adaptive resource management for component-based DRE systems. First, we describe the structure and functionality of the Resource Allocation and Control Engine (RACE), which is an open-source adaptive resource management framework built atop standards-based QoS-enabled component middleware. Second, we demonstrate the effectiveness of RACE in the context of a representative DRE system: NASA's Magnetospheric Multi-scale Mission system.

Many distributed real-time systems face the challenge of dynamically maximizing system utility and meeting strin- gent resource constraints in response to fluctuations in sys- tem workload. Thus, online adaptation must be adopted in face of workload changes in such systems. We present the MultiParametric Rate Adaptation (MPRA) algorithm for discrete rate adaptation in distributed real-time systems with end-to-end tasks. The key novelty and advantage of MPRA is that it can efficiently produce optimal solutions in response to workload variations such as dynamic task ar- rivals. Through offline preprocessing MPRA transforms an NP-hard utility optimization problem to the evaluation of a piecewise linear function of the CPU utilization. At run time MPRA produces optimal solutions by evaluating the function based on the CPU utilization. Analysis and simu- lation results show that MPRA maximizes system utility in the presence of varying workloads, while reducing the on- line computation complexity to polynomial time.