SCALABLE DISTRIBUTED FILE SYSTEM

Abstract

With demand for distributed system gaining momentum as a medium for information sharing, and existing systems not performing upto the mark when expanded to large scale, there is a demand for new type of distributed system that will withstand growth. The existing systems have not been designed to cater for large scale use. This dissertation is an attempt to achieve scalability and good performance in distributed systems. It Is also felt that incremental growth should minimally affect the central resources. At present system is primarily designed to be useful in academic and research type Institutions, and not for such applications like databases, where concurrent operations are excessive. The system, supports file system transparency. In order to achieve scalability caching, bulk data transfer, server initiated transactions, binary exponential timeout algorithm, simplified concurrency control, flexible location transparency has been adopted. Also set of Remote Procedure Calls (RPC) have been Implemented. In order to increase performance modified stop and wait protocol has been implemented. The file system has been built on top of existing two HP 9000/330 computers, supporting UNIX networked through Ethernet, with User Datagram Protocol (UDP) as the base. Multiple servers has been simulated on these two existing systems. The software is portable to any existing UNIX systems, which supports ARPANET services. Even though security and fault tolerance are very essential to highly scalable systems, the same has not been implemented, as the work involved is enormous