科学计算资源网格服务化研究与实践

Research and Practice of Gridifying Scientific Computing Resources

网格计算已经进入了以服务为中心的新阶段 ,共享的方式转为采用具有良好跨平台、互操作性的网格服务 针对科学计算资源 (主要是遗产型应用 )的网格服务化问题 ,进行了研究 ,提出了一个解决方案 :采用共同的抽象的调用接口来屏蔽各种各样科学计算应用接口 ;“按需”科学计算工厂解决了科学计算服务的管理与信息发布 ;统一的框架来监视计算的执行并进行运行时操纵 在设计方案的实现过程中 ,解决了底层包装和非阻塞调用问题 在原型实现的基础上 ,这个设计解决方案被成功地应用到了项目开发之中

A good many legacy applications are widely used in various scientific research fields.In pre-gridservice times, conventional means of sharing scientific computing resources is by sharing programs and data.With the introduction and wide adoption of OGSA framework and OGSI specification, grid computing has entered a new era of service-orientation, in which implementation of sharing domain expertise resorts to portable, interoperable grid services.So comes the challenge to gridify (to convert to OGSI-compliant grid services) existing scientific computing resources, which is mainly made up of legacy applications.Focusing on this topic, features of scientific legacy applications and key issues during gridification are analyzed, and resolutions are proposed, including a common abstract invocation interface for various applications, an 'on-demand' factory responsible for creating and publishing scientific computing services, and a uniform monitor and manipulation interface.The design scheme consists of three parts: scientific computing services which implement three interfaces of GridService, LegacyCodePortType and ManipulationPortType; scientific computing operation providers(OPs); “on-demand” scientific computing factories.During the course of implementation, a set of difficulties are resolved, including lower-level wrapping and non-blocking invocation.The design solution has been successfully applied to the development work of ChinaGrid project.Application scope tests show that the design solution can apply to most types of scientific computing legacy applications.Performance experiments show that the proposed gridification mechanisms bring about only slight efficiency loss.