Using Globus Grid Objects to Extend a Corba-based Object-Oriented System
Courtyard (room A)
Wednesday, 12:00, 45 minutes
Scott Spetka, ITT Industries and SUNY Institute of Technology
George Ramseyer, Air Force Research Laboratory
Richard Linderman, Air Force Research Laboratory
Demonstration number: 16Keywords: Grid, Services, Corba, CGI, CGICC, Web, Distributed, Framework High-Performance, Image, Inheritance, Polymorphic Abstract: The High-Performance Image Exploitation (HIE) FrameWork is an object-oriented system that is designed for management and processing of large images over geographically distributed high-performance computer (HPC) systems. The FrameWork client object implements a CGI user interface to Web browsers using CGICC objects for the client display. The FrameWork originally used Corba for inter-object communications with HPC-based objects. This demo addresses extending the system by incorporating Globus Grid objects. We added Globus Grid objects to the FrameWork to extend HPC access services. The Grid interface has significant features for controlling remote processing which enhance our distributed computing environment. These include: factory, discovery and notification services; a service browsing interface; and lifecycle management. Globus Grid objects extend the FrameWork to provide loosely coupled Globus services, in addition to tightly coupled Corba interfaces, to support flexible access to HPC systems. Object-oriented technologies, particularly inheritance, formed the foundation to implement a polymorphic net-centric approach to assure interoperability of our original Corba objects and Globus technologies-based objects and to accommodate future extensions. The demo will show how we use Corba and Globus objects to distribute HPC computations across multiple heterogeneous HPC systems using innovative scheduling algorithms. Object extensions, implemented using inheritance, allow our architecture to accommodate both technologies and also extend the system to support other distributed computing approaches, like RMI. This demo will show that our object-oriented architecture is the key to create a hybrid system that presents unique opportunities for parallel system developers to optimize performance for their codes.