Mutiprocessor Systems Scheduling For The Optimization Of Reliability And Energy Consumption

With the rapid development of VLSI technology and the great increase of application requirements, the multicore and multiprocessors architecture are already becoming more common and domain modern commercial processors. Multicore processor architecture is a highly parallel architecture and provides fast computing hardware platform. With the number of transistors on a chip increasing, its performance has been significantly improved. However, the improved performance is at the cost of a sharp increase in energy consumption of the chips. Energy consumption is one of the most important factors considered by many embedded real-time systems, especially for wireless and portable devices. In addition, with the microprocessor chip gradually adopt nanometer manufacturing process, the feature size of the integrated circuit continues to decrease, and its frequency continues to rise, which makes the failure rate of the processor cores continually raising and its reliability keeps decreasing. Hence, in the era of multiprocessor platform is widely used, how to schedule real-time tasks to meet the deadline, and to minimize energy consumption and to optimize the system reliability is a problem needs to be solved urgently in the field of system-level task scheduling. Solving this problem has important academic value as well as broad application prospects.Considering the multiprocessor systems and its urgent needs on energy reduction as well as reliable operation, this paper investigates scheduling strategies for energy minimization, which consider multiple constraints on real-time requirement and the reliability of the processors. The goal is to explore how much energy saving can be achieve by combining system-level scheduling methods with hardware energy saving techniques, and to explore the feasibility of the scheduling methods on optimizing the system reliability. The main contribution are listed as follows.(1) Considering the heterogeneity of reliability property of each processor, this paper solve the problem of energy-aware task assignment that guarantees to satisfy the real-time requirement and the reliability requirement. The algorithms that based on dynamic programming are proposed to produce optimal solutions for simple-path and tree-structured task graphs in polynomial time. When the input graph is a directed acyclic graph(DAG), the task assignment problem is NP-Complete. For the general task assignment problem for DAGs, we first develop an Integer Linear Programming(ILP) formulation to generate optimal scheduling results.Then, this paper propose a polynomial-time heuristic algorithm, to find near optimal solutions.(2) Technology scaling and ever increasing demand for performance have resulted in high power densities in current circuits, which directly results in a substantial increase in chip temperature. Owing to the increase in chip temperature, the lifetime reliability of the chip become increasingly severe and needs to be solved urgently. Existing work on system-level scheduling techniques, however, consider only thermal control or lifetime reliability optimization. This paper presents a Mixed Integer Linear Programming(MILP) model to determine the mapping and scheduling of real-time applications onto embedded multiprocessor platforms such that the total energy consumption is minimized while satisfying the lifetime constraint as well as the temperature threshold constraint. The proposed MILP model incorporates a lightweight thermal model that can predict the temperature more accurately and efficiently at design time.(3) In the big data era, resilient applications are found in numerous fields. By exploring this resilient property, the voltage over-scaling technique can be used on multiprocessor systems to reduce energy consumption. In this context, however, it is important to schedule computation tasks properly to maintain quality satisfied while ensuring that the overall system energy consumption is reduced, in the presence of random variations and errors incurred by VOS. This paper proposes a solid framework, namely ApproxMap, to determine the mapping and scaling sequence of resilient tasks to minimize the energy consumption of the application while meeting its quality requirements and timing constraints. When the task is resilient, ApproxMap trys to over-scale the processor voltage and use lightweight quality checkers to detect whether the output result is acceptable.