The Design And Implementation Of Manager Node System On GPU Cluster With Searching Function

Before the1990s, People Usually use Mainframe to complete high-performance computing. There are some advantages of Mainframe:focus application, easy to deploy, system monitoring, managing convenient. But Mainframe’s disadvantages are very obvious. There are some disadvantages:equipment are expensive, difficult to achieve high availability, operating systems and equipment are specialized, user Difficult to maintain by themselves, weak scalability. With the popularity of the PC Operating systems and the using of Intel CPU, People use Distributed System constituted by PC severs replace the mainframe. Distributed System solve a number of disadvantages of mainframe, but lost the benefits of mainframe. Cluster system combines the advantages of Mainframe and Distributed System. With the rapid development of the cluster system, Cluster managing system has been more and more attention and development widely.This paper describe the advantages and necessity of the cluster system firstly, describes the specific features and structure of GPU clusters with searching function nextly. Then describes the manager node’s position in this cluster, implement functional requirements analysis of manager node. The design and implementation of Manager Node System based on Win7/Win XP operating system of32bits, the role of Manager Node System is managing and monitoring the cluster system. The logical structure of manager Divided into three parts, the three parts are user interface layer, logic layer and data layer. The Interface layer generate interface that interact with the user. The Logic layer processing data information and background logical. The data layer including network I/O and local I/O data structure.This paper introduce a variety of techniques used by Manager Node System, describe the process of system implementation and test improvement lastly.