Challenges for Parallel I/O in Grid Computing
With virtually limitless resources, GRID computing has the potential to solve large-scale scientific problems that eclipse even applications that run on the largest computing clusters today. The architecture of a computing GRID simply consists of a heterogeneous network infrastructure connecting heterogeneous machines presumed to be larger than most clusters of the future. GRID computing uses heterogeneous resources to solve large-scale computational problems. With increasing dataset sizes in data-intensive GRID applications reaching terabytes and even petabytes, high-performance I/O is emerging as an important research area. The authors discuss much of the current status of research in GRID I/O. They also describe their research ideas for handling noncontiguous I/O access, consistency, caching, fault-tolerance and improved performance.