Research Of Software Algorithm Optimization Based On Multi-core Platform

With the development of semiconductor technology,processor manufacturers can integrate more and more cores on the chip.Multi-core processors have become the mainstream processors in current market,and integrate more cores in a single chip become a mainstream.Multi-core processor's execute behavior which is a build-in advantage of processor can provide higher performance at lower clock frequencies.Along with the innovation of hardware,people usually pose a higher expectation for software applications.However,the application layer of software calls the system hardware structure through the operating system software.So,consider the underlying hardware changes,how to use multi-core technology effectively is the main problem of multi-core software development.Multi-core software development method can be separated into multi-core operating system software development and multi-core application software development.Through approaches of heterogeneous operating system software scheduling optimization and multi-core platform application software optimization,this paper focuses on the optimization algorithms in three fields: improves the performance of the system,increase the utilization of the system and reduce the bottleneck of the system.The research includes: optimization of heterogeneous multi-core platform operating system software scheduling algorithm and multi-core platform application software optimization method.The following three subjects were research in this paper:1.A novel Match Type Scheduling Algorithm(MTSA)is proposed for the operating system on heterogeneous multi-core processor hardware.Through the integration of performance and functional asymmetry of the core in the heterogeneous architecture of multi-core processor,Traditional MTSA algorithm can use lower cost to achieve high performance and low performance thread execute concurrently,so it has value to research.But this structure of the operating system put forward higher requirements to software design;thread scheduling is one of the most critical issues.We propose a key indicator: match type factor(F),use it to describe the type of application.MTSA scheduling algorithm computed application of matching types factor F with dynamic monitoring sampling,find processor for the application.MTSA scheduling algorithm can dynamically adjust the core type that matches the application,perform thread migration scheduling,and maximize system throughput rate.Through the simulation on the experimental platform,results shown that the scheduling algorithm of type matching optimization can select the appropriate processing core by influencing the MTSA scheduling algorithm,so that the application can be allocated to the appropriate hardware to achieve the load balancing and improve the overall task speed,it can save time for the user.The performance improved about 8% time than other similar algorithms.2.A Parallel Optimization-Dijkstra(PO-Dijkstra)algorithm for multi-core platforms is proposed.With the development of Multi-core hardware,a lot of algorithm,especially algorithms with large amount of computation,obtain more development opportunities.In recent years,with the rapid increase of the vehicle,the upgrade of road construction and navigation system,Dijkstra algorithm has become one popular method of path planning.Based on graph theory,Dijkstra algorithm often used in the shortest path planning in the shortest path analysis application.However,due to the Dijkstra algorithm need traversal all the nodes and less efficient,it becomes the main bottleneck of practical application.The appearance of multi-core processors provides hardware foundation for parallel computing.Therefore,due to a large number of computation burden,Dijkstra shortest path planning algorithm has proposed a new research challenge for parallel processes.In this paper,we propose a parallel Dijkstra algorithm for multi-core platform(PO-Dijkstra algorithm).The traditional Dijkstra algorithm is split and parallelized by the multi-threaded programming tool Open MP for shared memory parallel system.Because thread splitting need to consume some resources,adaptive optimization will be carried out according to three parameters of total number of nodes K,the number of core M and CPU frequency.Experiments performed in the two systems,the total execution time of the PO-Dijkstra algorithm and the classic Dijkstra algorithm was studied comparatively.The experimental results show that: the speed of parallel optimization algorithm significantly increased,the maximum speed can be increased in 20-40%.3.Improved Particle Swarm Optimization Task Scheduling(IPSOTS)algorithm is proposed by optimized multimedia data transmission mechanism based on multi-core platform.With the development of the Internet,multimedia cloud computing platform development rapidly.However,due to the very large amount of data involved in multimedia resource,while the playback order and large-scale time requirement is also very strict,the increase of the process performance of multimedia data is very necessary.Multi-core processors can bring more benefit to multimedia data processing.Multi-core processor provides a stronger support to improve the processing performance of multimedia data.It has great study value to make full use of the bandwidth resources of nodes in multi-core environment and obtain data dynamic task scheduling for multimedia cloud computing platform.Based on the research of data dynamic task scheduling method of multimedia cloud computing platform,the system model and task model of dynamic scheduling of multimedia data are proposed.Based on the model,an IPSOTS algorithm based on particle swarm is proposed,through assigning the task optimization task scheduling strategy for processors can be obtained.The simulation results show that the method has high scheduling performance.Compared with neural network method,IPSOTS task scheduling algorithm shows comparative higher operation performance.It can shorten the delay time and increased of the system throughput rate in 15%.