Research And Application Of Tuplespace In Distributed And Parallel Computing

Performance improvement of single computer and advances in network technology brings to the field of high performance computing a new paradigm of computation—polymerizing network idle resources for large scale super computing. In this way, how to achieve the cooperation of dynamic, distributed computing resources, fulfill the solving of the complex computation problem is an urgent need.Based on a comprehensive study of cooperation technology in distributed computing technology, through analysis and weigh, choose the tuplespace for coordination between computing resources according to the actual demand. The task cooperation mechanism based on generative communication in tuplespace was deeply researched. We fully utilized the spatially uncoupled and temporally uncoupled characteristics of generative communication, selected the Master-Worker model to design the task cooperation system based on tuplespace. The whole design follows the principal of SoC (Separation of concerns), separates the data transfer functions which the user interested from tuplespace management functions. We designed the parallel programming function library and tuplespace manage subsystem respectively.Concrete realization applies hierarchical encapsulation strategy provided users with a parallel programming library. In tuplespace management subsystem, on the communication bottle-neck that exits in centralized access of single tuplespace, we proposed a multi-layered structure of tuplespace under the centralized control; through building multi-tuplespace and data distribution decentralized communication is realized. In tuplespace storage management we use virtual memory management of OS for reference, swap data between inner and outer memory and hide the actual capacity of tuplespace. Finally, check point and access control was used for fault tolerance and security of the system. Test results demonstrated that the coordination system realized the desired goals and achieved a good speedup for the enumeration type problems.