Research On Task Migration And Scheduling Algorithm For Heterogeneous Multi-core Processors

With the development of multi-core processors,the process scheduling and task migration have become one of the hot research topics in the field of high performance computing.At present,the multi-core process scheduling algorithm and task migration algorithm are still not ideal,there are many problems such as inefficient allocation efficiency,high dynamic algorithm complexity,load imbalance and so on,which seriously affect the performance of multi-core processors.Tasks will show different operating characteristics and different hardware resource requirements.Compared with homogeneous multi-core processors,heterogeneous multi-core processors can improve the system throughput and reduce processor energy consumption.In the Linux operating system,the scheduler in dealing with user level tasks without considering the contact between the tasks and the characteristics of each task.In this paper,we study the scheduling algorithm and migration algorithm for heterogeneous multi-core processors,and propose a Cache-IPC aware task migration algorithm which is suitable for heterogeneous multi-core processors.On the basis of analyzing the scheduling mechanism and load balancing mechanism of linux2.6.34 kernel,the improved process scheduling algorithm based on Linux kernel is realized.Cache-IPC sensing task migration algorithm is implemented by using PAPI driver in user space to obtain the process characteristics of the processor related event register.The migration algorithm consider the contention of the processes on the shared cache when the process runs on the adjacent cores,let the process with high MPKI(Mispredicts per kilo instructions)insert one queue and let the process with low MPKI insert the other queue,so as to ensure the process with and the process with low MPKI run on the two adjacent cores,so as to enable the process with high MPKI and process with low MPKI run on the adjacent two cores.After the first classification of the process,we let the processes with high IPC run on the large core,let the rest of the processes run on the small core.Experiments show that compared with the Linux CFS scheduling algorithm,the performance of the Cache-IPC aware heterogeneous multi-core task migration algorithm is improved by 8.65%.The improved scheduling algorithm which is based on the Linux kernel,on the basis of entirely fair scheduling algorithm,fully consider the process characteristics of the runtime and the differences of load between the cores.The algorithm uses the cache features of the process to cluster so as to reduce the shared cache contention.By analyzing the characteristics of the task instructions,the processes is divided into cup-bound state and memory-bound state.We propose a new migration factor M(M = x×P,P:processor execution ability,x:proportional coefficient)which is both suitable for big cores and small cores.According to the process execution state for scheduling between big cores and small cores.At the same time,in order to adapt to the heterogeneous multi-core environment,the calculation method of the task load is modified according to the computing power of the cores.Experiments show that compared with the Linux CFS scheduling algorithm the performance of the process scheduling algorithm based on the Linux kernel is improved by 9.81%.