Research On Technology Of Managing Shared Memory In Multicore Operating Systems

With the development of technology of microprocessors and the improvement of the demand for computing abilities, MPSoC has become a major trend of processor development. The appearance of multiprocessor systems makes the competition of shared resources more fiercely. It makes the efficiency of managing shared resources having great influences on the performance of the system.Memory system is composed of some storage arrays (bank) which can be executed in parallel. How to use banks efficiently decides the performance of the system. The delay of accessing memory is affected not only by the restriction of hardware, but also by the parallel execution of memory accessing requests and the probability of bank conflicts. However, traditional methods of optimizing memory can’t solve multi-tasks competing with shared memory leading to the poor efficiency of memory accessing on multicores.In this problem, this paper devises a method of managing memory with multiprocessors accessing simultaneity on operating systems. The main idea is to add policy of perception of memory to paging paths, which is in order to realize the partition of memory space and makes the task accessing memory in a controllable manner. Paging method in current operating system choose physical addresses for a process randomly, so memory space of a process is uncontrolled. Different processes in different cores will interfere with each other mostly. Adopting the method in our paper, it is efficient for different processes to choose its banks to reduce such conflicts.Some people in academics adopt a method based on paging in operating system to manage shared cache accesses in multiprocessors to reduce conflicts caused by different processes mappings. Our research is a complement and improvement of current paging methods of managing cache, which is taking cache mappings and memory mappings into consideration. In the meanwhile, with processes in different cores have different priority, our thesis adds methods of management of priority-driven based bank requests. In order to validate our policy’s correctness, our thesis do experiments on full-system simulator Simics and GEMS.From the results of experiments, we can see our method can solve the problem of the increasing execution time due to task conflicts and the unpredictable execution time, which can be very practical.