Research And Implementation On Task Schedule And Resources Management Technique Of High Performance Parallel Visualization Server

The technology of scientific computing visualization can transform scientific data which researchers could not understand directly into the image or visual sense information which people can sense with eyes intuitively. Recently,it already has become an indispensable technique or tool in the fields of scientific computing and numerical simulation. It is also applied significantly in the fields of biomedicine, computational fluid dynamics, atmospheric physics, geophysical, aerospace and other scientific research as well as defense and national economy. With the continuous development of high performance computing technology, the ability of the scientific computing improves continuously. The data scale generated by scientific computing increases form GB level, TB level to PB level, even to EB level. The data scale gradually becomes bigger and bigger and the physical phenomena and law in the data become more and more complex, resulting in that the scientific visualization technology faces many new technical challenges.Building high-performance parallel visualization servers based on super computers can take advantage of the computing and storage resources of high-performance computers and overcome the time-consuming importing or exporting data of traditional post-processing parallel visual model and the difficulty to implement steering visualization. And then it will effectively improve the efficiency and quality of the large-scale scientific data parallel visualization, and has become an important development direction of the high-performance parallel visualization. Because of the needs of building high-performance parallel visualization servers, this study does a lot of research focusing on hardware and software architecture framework, visualization server task scheduling, resource management algorithms, and other key technologies of high-performance parallel visualization server. The main work and research achieved are as follows:1.Put forward a solution to build the hardware and software system structure of high-performance parallel visualization server based on supercomputer. Hardware system structure occupies a part of high-performance computer node resources relatively independently,which can be exclusively used in data storage,I/O communication, graphic plotting and resource management. This method will not only guarantee the independence of the parallel visualization server hardware resources relatively, but also ensure the efficiency of the parallel visualization well. The software system architecture take full advantage of the existing resource management software of the high-performance computers, according to the requirements of parallel visualization and multi-user remote interactive mission, and then extends a visualization task scheduling and resource allocation module, which will not affect the stability of the existing high-performance computer resource management system, but also ensure the high efficiency of the parallel visualization tasks and effectiveness of multi-user remote interactive visualization.2.In view of the characteristics and needs of the service object’s jobs, put forward a task scheduling and resource management algorithm based on the task properties selection. By defining the task properties abstractly and the parallelism of the tasks quantifiably, allocate server computing and storage nodes resources reasonably, so as to make sure that the task which needs acceleration by parallel computing urgently gets the resources preferentially. The results of the simulative experiment show that the algorithm can effectively improve the user experience and task execution efficiency.3.In order to ensure the effectiveness and correction of the task scheduling and resource management algorithm based on the task properties selection, we could come up with a task properties maintaining method based on linear regression model. The method takes advantage of the linear regression model in the modern engineering mathematics, and makes use of the information of the applications’ running history in the server and relevant statistical information, to complete maintenance and update of the attribute of the application algorithm. The results of the simulative experiment show that this method can improve the efficiency and accuracy of the task scheduling and resource management.4.Making use of the existing laboratory conditions, we could develop a prototype test system of high-performance parallel visualization. Users can visit the prototype test system of high-performance parallel visualization remotely via web browser by laptops, cell phones and other devices. And we use this prototype system to visualize some typical numerical simulation data instances of flow field, and achieve some better results, which proves well the correctness and effectiveness of this study.