Issues in distributed operating system support for distributed database managemen

Organizations are becoming richer in computer hardware as it becomes cheaper and more powerful. At the same time they are accumulating unprecedented amounts of data. Some organizations buy computers specifically designed for database functions. Others accumulate database processing potential through addition of general purpose workstations to service specific functions for their employees. In the future, we anticipate that knowledge based front ends to conventionally structured databases will become increasingly important. The queries which can be expected from such an environment may be significantly more complex than the simple queries that are the mainstay of database technology today.;In the past decade, distributed operating systems (DOS) have been developed to facilitate access to shared resources in networks of computer systems. The revolution in main memory technology has resulted in great increases in work station memory capacity. Along with a reduction in cost for increasingly powerful processors, this has made networks of relatively powerful machines quite common. In spite of significant advances in hardware technology, software has lagged behind in providing operating system functions capable of harnessing the potential for improved performance. Memory added to systems has been used primarily for support of virtual memory operation. Greater amounts of main memory can more efficiently support increases in size for program virtual address spaces and the system I/O buffer pool.;The focus of our research is distributed database management systems (DDBMS). We describe DOS facilities that are available in current systems and how they fail to efficiently support the I/O requirements of advanced DDBMS. We explore the potential for improved performance in database query processing for algorithms that can take advantage of large memory. We introduce hash join query processing techniques and describe a new algorithm for processing 3-way join queries. Query optimization depends on knowledge of memory availability for algorithm selection. We propose DOS functionality designed to reduce I/O cost and provide improved support for parallel processing. We use an experimental DDBMS implementation to quantify the potential for improvement from a DOS implementation of our proposed functionality.