Database systems: Making UNIX∗ operating systems safe for databases

The UNIX∗ operating system was written for document preparation and software development. Its general utility has led to its use for many other things, including database management. There have been complaints that the system kernel is unsuitable for running applications that use databases: the file system is accused of being too slow and unrobust; the process structure is accused of fatal inefficiencies; and there is neither enough sharing nor any way of controlling what there is. Modest changes to the kernel refute the attack. The file system is robust and can be made much faster; in 32-bit paged systems the process structure matches the needs of transaction processing systems, and a few additional system calls provide all that is needed for sharing and its control. This paper discusses the problems and shows how to solve them.