Research On Parallel Scheduling Method For Rasterized Distributed Hydrological Simulation

To meet the expansion from population to environmental pollution,climate change and water shortage and other practical needs,the development of hydrological science is the healthy development of China's national economy and a solid foundation and protection.Facing the hydrological problems of major disciplines,distributed hydrological simulation of large-scale watersheds has a very important role and status.However,when the distributed hydrological simulation of large-scale watershed high-resolution,long-duration and multi-geographical coupling process is large,the data-intensive computing problem is highlighted,and the task scheduling and resource allocation have some blindness.In the existing research in this field,the different types of watersheds adopt different levels of parallel computing method.When the simulation unit is singular,the parallel efficiency is limited.In short,there are two key issues to be addressed,namely: the lack of effective distributed hydrological model parallel algorithms,and the lack of task scheduling and resource coordination mechanism.First,we analyze the parallelism of distributed hydrological models and compare the feasibility of their respective implementation.Then,we propose a distributed hydrological model framework based on the principle of ecological optimality by referring to the double-layer discrete decomposition of space-time domain.So this study is for this distributed hydrological model of the parallel computing framework to start,it mainly contains the followings:Firstly,the computationally intensive or data-intensive features of the task-dependent and parallel processing of the coupling is too high,we use the DAG scheduling parallel method to deal with inter-task dependencies.Secondly,in the DAG scheduling algorithm,the time and space complexity of the table scheduling algorithm are well performed,but the quality of the solution space is very low.Random search algorithms such as genetic algorithm and simulated annealing algorithm have excellent solution space,but its task scheduling overhead is relatively large,even for different DAG map has a different control parameters,so itis difficult to achieve.Therefore,on this basis,taking into account the two aspects of task scheduling overhead and parallel efficiency improvement,we weight and compromise,propose a task based on the height of the DAG scheduling algorithm to dynamically adjust the task priority.Thirdly,the task and the resource match are blind in solving the problem of intensive computing task throughput.In this paper,we propose an adaptive resource common allocation model,which fully considers the resource performance measurement and task calculation density,and realizes task scheduling and resource adaptive matching.The experimental results show the effectiveness of the parallel algorithm and the validity of the adaptive resource coordination model.The method has good robustness and expansibility.