An object-oriented framework for efficient data access in data intensive computing

Efficient data access is important to achieve high performance in data intensive computing. This paper presents a method of passive data access in the framework of ParoC++-a parallel object-oriented programming environment. ParoC++ extends C++ to distributed environments with the integration of user requirements into parallel objects. Passive data access enables thedata source to initiate and store data directly to a user-specified address space. This ability allows better overlapping between computation and communication by data prediction, partial data processing and auto-data aggregation from multiple sources. Some experiments have been done, showing the scalability and the efficiency of passive data access in ParoC++ compared to direct data access methods.