Why Parallel File Systems?
A Parallel file system has primarily evolved to cater to the requirements of scientific and engineering application community where large IO throughput is required for a single application. In a simpler model, this requires every single IO operation to be handled by large number of disk spindles to provide high throughput. Also, the threads in the application job typically share a set of few files in a complex, yet well-defined manner. Issues are further complicated for a distributed memory parallel machine where the data sets are scattered across the different application threads on various nodes require sophisticated scatter-gather communication operations. Design issues of the parallel file systems address these requirements by providing a parallel IO programming interface and also a system architecture that can support high throughput. With clusters emerging as an affordable parallel computing platform, commodity based PFS implementation such as C-PFS provide high performance IO on
