Research On Resource Reconfiguration And Task Scheduling In General Heterogeneous Computing System

In this era of rapid development of information technology,traditional computing systems with the core of central programming unit seems to be struggling when facing massive data.With the emergence of multiple encoding methods and communication systems,there is an urgent need for a new type of computing system that integrates multiple types of computing resources to meet the growing computing needs in multiple fields.The general heterogeneous computing system contains a wealth of computing processing resources,which fits well with the current requirements for new computing systems.In this thesis,based on general heterogeneous computing system project,current resource reconfigurable problems are summarized and the theory of task scheduling is elaborated as well,aiming at the problem of how to improve the utilization rate of system resources and scheduling efficiency of the system.And on the base of that,the problem of resource reconfiguration and task scheduling is taken for research in general heterogeneous computing system.In order to realize resource reconfigurability,first of all,from the viewpoint of openness and modularity,decouple the hardware and software of general heterogeneous computing system,and then build a hierarchical software and hardware architecture.Then based on the system architecture,design a componentized software resource reconstruction plan,formulate component specifications,clarify component models and component reconstruction methods,and construct componentized computing processing procedures.The reconfiguration of system software resources will cause the dynamic changes of system hardware resource functions.In order to solve this problem,three different FPGA reconfigurable schemes are designed and the specific implementation process of the schemes are analyzed,taking FPGA computing resources as an example.The results of the experimental above show that those three reconfigurable schemes all can work well.To improve the computing efficiency of the system,it is necessary to design a task scheduling algorithm with good performance.A multi-level task scheduling algorithm with restricted tasks is proposed in this thesis,after clarifying the system resource reconfiguration scheme.But in general heterogeneous computing system,different computing processing procedures will lead to the functional reconstruction of the system's computing resources.However,traditional list scheduling algorithm does not consider the time consumption caused by functional reconstruction.So as to avoid additional configuration time consumption,the concept of restricted tasks is given in this thesis.The choice of task computing nodes will be directly affected by restricted task whose meaning is that there is a certain dependency relationship between tasks and computing nodes.Restricted tasks are joined as constraint to the phase of choosing computing node.Subsequently,new principles about matching computing node are formulated.Meanwhile,the method of calculating task priority is improved to obtain a more reasonable task scheduling queue.Finally,multigroup contrast experiments are designed to compare and analyze proposed algorithm with other scheduling algorithms from multiple perspectives.The simulation and experimental results prove that in the general heterogeneous computing system scenario,the proposed method has the following notable features,that it needs brief time to complete task scheduling,the relative scheduling length are short,the task scheduling efficiency is high,and it has a strong actual use value.