Abstract
Advances in technology lead to remarkable performance improvements in computer systems nowadays. Personal computers (PC) equipped with high speed network interfaces are affordable and easy to acquire for researchers. A group of loosely coupled computers that work together closely form a cluster. PC clusters have been commonly used these days to solve CPU-intensive tasks. But the speed improvement of hard disks can not catch up with other components in the system. The storage subsystem becomes a bottleneck for I/O-intensive tasks. Therefore, parallel file system is widely used in clusters to provide high performance I/O.Microsoft Windows operating systems share an enormous portion of the PC operating system market. But most of the existing parallel file systems are based on UNIX-like operating systems. We implement a parallel file system for Windows using the Microsoft .NET framework. A .NET library is provided for users to achieve high performance I/O by using our parallel file system. In addition, a file system driver is implemented to trap Win32 API I/O function calls. Thus existing binaries can access files stored on our parallel file system without recompiling. In this thesis, the design and implementation of our system are described and the performance is also evaluated. The results show that the write performance reaches a peak of 110MB/s and the read performance reaches a peak of 75MB/s.