| The concept of grid computing is gaining polarity with the emergence of the Internet as a medium for global communication and the wide spread availability of powerful computers and networks as low-cost commodity components. The computing resources and special class of scientific devices or instruments are located across various organizations around the globe. These resources could be computational systems,special class of devices,visualization platforms,and storage devices. A number of applications need more computing power than can be offered by a single resource of organization in order to solve them within a feasible/reasonable time and cost. This promoted the exploration of logically coupling geographically distributed high-end computational resources and using them for solving large-scale problems. Such emerging infrastructure is called computational grid.Grid can be called the third Internet. It tries to connect all resources of Internet and make Internet a large virtual machine,and share computational resources,storage resources,communicational resources,software resources,informational resources and knowledge resources. The grid has widely application area,includes distributed supercomputing,high-throughput computing,on-demand computing,data-intensive computing and collaborative computing. So the research of grid and relatively area is very important.Our research project started from 1998,the goal of which is to develop a practical Grid Management System with single system image named JobCenter-Grid. Through division management model,the management of system includes cluster management and node group management,and the system supports site autonomy,heterogeneous substrate,node extensibility and transparence. We research the implementation of single system image in grid,and put forward the implementation method. At the same time,we discuss the scheduling of job and resource from two folds:with and without resource requirements. The main works and achievements of the authorsince 1998 cover the following aspects.1. Put forward the concept of single system image in grid. Its implementation method includes application layer and system layer. Application layer includes single entry point and single control point,and it can make user see a single system. System layer includes single user management,single resource management and single job management system,and it can offer the service for application layer.2. Put forward the idea of division and management for the grid. The grid is composed by many clusters,and the cluster is divided to many node groups with centralized management. Cluster and cluster are peer-to-peer with decentralized management. It can realize node autonomy,and avoid management neck,and enforce the efficiency.3. Employ IP address dynamically mapping method to implement single entry point. The implementation of single control point includes single monitor with PULL model and single control with PUSH model. Then we put forward hierarchical control structure to implement single control point.4. Put forward the account management model based on lazy consistency protocol. On the basis of this model,employ the unified password authentication to identify the user. In order to implement high efficiency,security and flexibility,we use temporary account to access the remote cluster.5. Put forward decentralized hierarchical resource management model. On the basis of this model,put forward a resource-matching model of "dark match-ing-confirm-exact matching". We design the global resource scheduling algorithm and cluster resource scheduling algorithm to distribute resource effectively.6. Design the complex job network. Then according to the character of job network,we design the advance-scheduling algorithm based on barrier to schedule job network to shorten the waiting time. And we also research on the job without resource requirement,and design proportional job scheduling algorithm. We also design the adaptive method of job execution,and through which the resource can...
|
PDF Full Text Request