Scheduling, memory access and system reliability optimizations on multiprocessor architectures

Edwin Sha,Yi He

2011-01-01

Abstract:With the fast development of the processor technology, the processing units in systems are becoming more and more powerful. Meanwhile, however, the single-core processor development is hitting major physical barriers and the memory wall. A practical solution to this problem is putting more cores in a system, so that computations can be parallelized as much as possible. To overcome the physical barriers and the memory wall of the single-core processors, MPSoC systems are more frequently deployed. The bottleneck for improving the system performance for these kind of systems often turns out to be the need to make memory accesses by multiple cores more efficient. To improve the system and memory utilization, the key is to take advantage of locality as much as possible and integrate it into task scheduling. In the first part of this dissertation, the problem of co-optimizing the memory accesses and task scheduling of an MPSoC with multi-level of memory is discussed. A variety of system architectures and memory hierarchies are explored, with the goal to co-optimize the performance of the memory access and task scheduling. In each architecture, the definitions of the problems and the system architecture of the systems are discussed. For each problem, several algorithms are proposed. Integer linear programming techniques are used to achieve the optimum results in these problems. Some heuristic algorithms are also proposed to get close to optimum results with polynomial running time. Experimental results show the simulation results by implementing the proposed algorithms and the analysis of the results is also presented. In the second part of this dissertation, the focus of the research centers on the reliability and heterogeneity of multi-core systems. In recent years, more and more heterogeneous processor cores are embedded into a single chip. An important research problem is to maximize system reliability while satisfying the required time constraint. A scheduling scheme is needed to exploit the heterogeneity of a system and satisfy both the reliability requirement and the given time constraint. This kind of heterogeneous systems are implemented in a wide variety of applications, especially real-time embedded systems. For these problems, Integer linear programming techniques are also used to achieve the optimum results. Several heuristic algorithms are proposed to achieve good results with polynomial running time. A comparison between our algorithm and some existing algorithms is also presented.