Design And Research Of A Computing Grid Prototype System

With the rapid development and wide application of Internet technologies, a new computing pattern, Grid computing, which substantially changes the way we perceive and use computational resources and services is booming. Grid is an Internet-based infrastructure for resource shareing and cooperative resource, which brings the Internet into a new era. Grid is a compositive computing and resource environment, which can contain all kinds of computing resources, and change them to a environment which has reliable, standard and ecomonic computing capability. Besides the all types computer, the reousrces include network, data, apparatus and people, etc. Grid research is in the exploring phase yet, which includes computing grid, data grid, information grid and simulating grid, etc. Computing grid can be able to aggregate the homogeneity or heterogeneity computer, workstation, cluster (NOW and COW), database, apparatus and storage device, etc, and change them into relativily transparent, virtual high performance environment. Grid computing inherits the character of the paralleling, distributed system, but because the resources distribute in wide area, and has the characteristic of heterogeneity and dynamic, it must confront many challenging technic problems. Because the research of computing grid begins early, and its infrastruture and facility's study have been already mature, our research focuses on computing grid. We have developed and studied the agent-based Campus Computing Grid System.The reason of putting forward the studying is there have contradiction between the high performance computer which is not used all the time and the researcher who has not enough resource to use: on one hand, many researcher want to improve the research level urgently with the computing resource help, but their own's apparatus cann't meet their needs, and so they cann't developing reseach well; on the other hand, many high performance computer and PC are not used in our campus, because they are not able to find enough computing application. The reason is there has an interval between the researcher and the high performance computer, because researcher understands deficiently of the high performance computer and has difficult to use the high performance computer. Our research wants to provide a grid platform to provide computing service such as paralleling computing, and its goal is to couple up the dispersed computing resource into a cluster which has high computing capability. Now the system primarily uses our lab's resource, including PII/266, PIV/2.4G, PIV/2.6, DISC Array, etc. to provide cheap computing service. The system provides user with web accessing interface, and the user can use computing resource through browser in anywhere; they can submit task in distance, and query the status or result of the task; they can analyze the processor, memory's availability and network spending while the task is running; user can easily get result of their job. Computing Grid Prototype System includes three parts primarily: grid client-side, the agent-based middleware, grid server-side. Grid client-side is to resolve the problem that user want to use all kinds of resource through uniform web portal. The portal includes the modules of register and login, job submitting, job executing, result and status querying, information querying, etc. If user login our grid portal, they will found that using complex grid resource is easily like browsing web page. Grid server-side takes charge of executing the user's task. Each computer has been installed Globus Toolkit 2.4, and established valid authenticated grid user whose accessing right is limited. Through installing and configuring MPICH-G2 on all computers, we construct a cluster which can provide high performance computing service. Grid server-side does not provide accessing interface to user directly. The resource is transparent to user and user doesn't need to consider the place of the grid resource. The user only need to submit their job in the portal and all the other work is be done by the agent which is an agent-based middleware, which takes charge of communicating with server to help executing user's job. The agnet-based middleware has four modules: job scheduling, job executing, information service and grid databases. Job scheduling module takes charge of allocating resource and scheduling job. Our job scheduling strategy is first coming first servering. After user submit their job and does not specify the appointed computing node, the scheduling module will query the information of the system resource, and if there have the nodes whose resource informations can saticify the need of the user's job, then, the job will be submitted to the nodes. After that, the executing module will be activated to run the user's job. Job executing module includes two parts: compiling and running program. The processes whose work is compiling and running program are waiting for message of the scheduling module on the nodes at all times. If receiveing the commond of executing job, the process will run the program with the user's parameters. When finishing running the job, the result will be sent to the scheduling module, which will dispaly the result to the user. Information service module provides the information inquireing. We construct grid resource information index by configuring the GRIS and GIIS on all of the grid nodes. User and the agent can acquire the resource information by accessing the information server which is implemented through the CoG Kit. Campus Grid Computing System use the mysql to be the grid databases which store the information of the user and the job, and it connect the user and the agent to communicate. Computing Grid Prototype System views the "page computing"as the basic...