The Research On Dynamic Task Management And Scheduling Algorithm Based On Grid Simulation Computing Environment

Computer simulation technology is an important means and method of analyzing and studying systems operation behaviors and revealing the dynamic movement process and principles. In recent years, with the development of in-depth research of systematic science, the control theory, calculation technology and computer science and technology, computer simulation technology has become a new discipline. However, due to such "bottleneck" factors as lack of computing capability, limitation of network bandwidth, un-compatible of protocol and the network security, to some extent, the efficiency of the practical simulation application is undermined, and also the depth and width are limited. Especially for compute-intensive, data-intensive and lots of interaction with the remote control complex simulation, using the existing simulation technology and computing infrastructure have been difficult to achieve or inefficient.At present, the characteristic problems which the traditional simulation faces to are listed as follows:(1)In the compute-intensive simulation, laboratory has no high-performance computers, therefore, must bring in a few computers, and a machine, for computing a complex federal member, needs a few days or even a few weeks to complete a simulation task.(2)The process based on the inference of simulation always needs to be executed distributed, and consequently, more PC need to be collected, because when the common PC are engaging in the computing-intensive simulation, they have no the extra capability to finish the daily work. On the other hand, at night or non-working days, a large number of machines are in the idle state, and such a contradiction exists in most of companies. Therefore, resolving this problem based on grid has the great practical significance.Grid technology can make up for deficiencies in the fields mentioned above, and in order to solve the problems, this paper presents a grid-based Integrated Simulation Framework GFS (Grid-based Framework for simulation). The idea of GFS is to improve the resource management efficiency in the grid simulation from two levels: simulation methods and software design; Simulation). GFS idea is the simulation methods and software design of the two-level increase in the simulation grid resource management efficiency: from the aspects of methods, the simulation development process, simulation tasks submission and the configuration methods are improved, and at the same time, various optimization algorithms are used to select Grid resources and the implement task scheduling; in software design the grid (Grid Service) protocol standards are used to modify simulators and correspondent supporting components, and provide for users the transparent operation of the simulation and task management services.Overall, the GFS framework has two kinds of capabilities: Simulation task management and integration capabilities.Simulation task management capabilities are mainly concerned about needs of simulation execution towards computing and communications resources, including the following features:(1)It can effectively support transplantation from the traditional modeling and simulation software environment to grid environment. If we only consider the simulators left are transplanted to be executed under the grid, and do not consider such problems as the efficient resource management, then this is a problem which has been solved under the traditional network environment, because people in the WAN or Internet has been launched a large number of simulation experiments. The emphasis of transplantation does not lie in "how to transplant", but in how to utilize the grid resource management advantages after transplantation. GFS must play a role of balancing grid system and traditional modeling simulation environment.(2)It can integrate the existing grid middleware and related standard components. Generally the relatively mature grid middleware is mainly of Globus Toolkits series, because it is the realization of the de facto standard for Grid OGSA, OGSI, which has been widely used; other related components mainly refer to grid resource management tools, such as resource management client terminals tools Condor-G, Nimrod-G, as well as grid resources enquiries, forecasting tools such as NWS (Network Weather Service) and so on.(3)It can support the sequence simulation tasks. Generally sequence simulation can be naturally divided into parallel and non-dependent relationship sub-tasks, which is the so-called EP mode.(4)The task dynamic scheduling. Dynamic Scheduling includes selection of suitable grid nodes according to the task needs and assignment of tasks to the grid nodes to implement by scheduling strategy. Support the dynamic simulation run-time task scheduling. Dynamic Scheduling mainly considers load balancing and the decrease of communication delay.(5)It can support reusability of modeling and simulation. General modeling and simulation activity is a standardized process, such as: whether to develop federal according to HLA must follow certain steps. Therefore, the majority of modeling and simulation activities have reusability, and moreover, task management should support this reusability from the aspect of software.Grid is a computing environment. Existing resources are the bottom facilities for grid computing. Computing applications are called through grid to share online resources and complete the task. For the reason of the distribution and heterogeneity grid bottom equipment, the task management, task scheduling and resource management in parallel tasks are the difficulties and priorities for research.The fundamental purpose of parallel is to accelerate. People hope that n nodes in the parallel system can run faster n time than single computer. This usually cannot be achieved in reality, and even very different. Obviously, nodes for parallel computing cannot give full play to their capability. This situation is even more serious, when more structure of applications is relatively complicated. Therefore, in parallel processing system, how to improve the utilization rate of the various nodes is the key to impact the computing performance. In practice there is often such a situation: When some of the nodes are busy with computing, but others are of idle state.This, of course, greatly reduces the utilization of nodes. The load balance which is studied in this paper is an important way to improve the parallel performance at present.The basic objective of load balancing is to distribute operation to various nodes through task scheduling,, thereby enhancing system resources (mainly CPU) utilization. This paper discusses and designs a non-pre-empt dynamic scheduling strategy. First this paper analyzes the advantages and disadvantages of non-pre-empt scheduling in the cluster system, as well as the characteristics of dynamic scheduling and dispatching system and next introduces the important algorithm used in the task scheduling, finally designs and implements a non-pre-empt mpi-based load balancing system, and gives performance evaluation.In dynamic parallel scheduling, non-first approach, due to its features of easy implementation, wide range of application and flexibility is widely used. But it also, has some fundamental flaws, such as the distribution of decision-making can not be changed, and in parallel grid system environment, the operation of the dynamic scheduling system itself has also brought additional costs. This paper designs and implements a non-pre-empt load balancing system based on PVM.The system is centralized load balancing system, and because the regulation algorithm of the centralized system is simple and effective, centralized management is a relatively good choice to improve the performance and reduce the expenditure. Expenses mainly include: load information collection expenses, the exchange of information and scheduling expenses. Among them, the collection of load information adopts a self-adaptive dynamic regulation of the time interval algorithm, to ensure that the more large volume of parallel computing tasks, the smaller the introduction of the additional overhead, and the exchange of information only changes in the larger load circumstances. The system cost is relatively small, and in the actual test, the single computer running time of loading application does not exceed the normal scope of fluctuation.