A remote file system for heterogeneous network topologies

Achieving high performance in a distributed computing environment is a combination of efforts between the network, distributed operating systems, and the applications. All too often the focus of improvement excludes the opportunity for the application to make an impact on performance. In this study we provide a facility where the application characteristics can be employed in formulating a distributed solution. The specific problem being considered is remote file services, although the general ideas can be applied to other areas. Our file system permits the application to indicate the type of caching and transfer strategies most suitable for each remote file access instance: depending on the application will effect the load realized on the network. Earlier work provided techniques to characterize application behavior through simple profiles. This profile data can be employed in selecting what kind of caching and transfer strategies would be most suitable to minimize traffic on the network. We describe the architecture of our file system. We show how the system provides for efficient network performance by realization of reduced traffic on the network for each application. This application data approach will always provide better performance than any random based model design