Collaborative Research: CSR---AES: InterGridSolve: A Virtualized, General Purpose, and Interoperable Grid Computing Environment for Computational Science

InterGridSolve项目正在开发一种环境，使科学家能够轻松利用由计算服务提供商(CSP)提供的网格资源，这些计算服务提供商(CSP)的范围从联邦资助的TeraGrid到商业可用的服务，如Amazon Elastic Compute Cloud(EC2)提供的服务。为了支持许多科学家在工作中使用的科学计算环境(SCES)，并结合灵活的计算模型，该研究正在增强GridSolve，这是一个预先存在的代理RPC环境。GridSolve基于开放网格论坛(OGF)提出的GridRPC API，使用函数句柄和会话对网格资源进行远程过程调用。GridSolve包括资源调度、执行监控和容错。正在设计通用数据移动机制(即数据句柄)，并将其添加到GridSolve中，以便启用工作流应用程序。数据处理机制将被建议作为GridRPC API的扩展。正在探索使用GridSolve中的数据句柄在网格资源上启用工作流应用程序的工具，并正在实现几类工作流应用程序(如简单的DAG)。更广泛的影响：该项目将使多个领域的计算科学家能够使用他们习惯的SCE来访问由各种服务提供商提供的网格资源，并在这些资源上运行计算密集型的工作流作业。

The InterGridSolve project is developing an environment that will enable scientists to easily harness Grid resources provided by Computational Services Providers (CSPs) ranging from federally funded TeraGrid to commercially available services such as those provided by Amazon Elastic Compute Cloud (EC2). In order to support the Scientific Computing Environments (SCEs) that many scientists use in their work (e.g., Matlab, Octave), and to incorporate flexible computing models, the research is enhancing GridSolve, a pre-existing brokered RPC environment. GridSolve is based on the GridRPC API proposed by the Open Grid Forum (OGF) and uses function handles and sessions to make remote procedure calls on Grid resources. GridSolve includes resource scheduling, execution monitoring and fault tolerance. General purpose data movement mechanisms (i.e., data handles) are being designed and added to GridSolve in order to enable workflow applications. The data handle mechanisms are to be proposed as extensions to the GridRPC API. Tools to enable workflow applications on Grid resources using data handles in GridSolve are being explored, and several classes of workflow applications (such as simple DAGs) are being implemented. Broader Impact: This project will enable computational scientists across multiple domains to use their accustomed SCEs to access Grid resources made available by various service providers, and to run computationally intensive, workflow jobs on these resources.