Dynamic program updating in a distributed computer system.

Throughout its lifetime, a given computer program will continue to evolve as enhancements are added and bugs are detected and removed. In some circumstances, interrupting old versions of a program to load the newest version is impractical due to the high cost of doing so. This dissertation describes a method of loading and running a new version of a program without stopping the old version (dynamic updating). Previous work on this problem imposed significant programming constraints or caused substantial performance degradation of the program being updated. This work addresses these problems by providing a transparent environment to the programmer using the system. A conceptual overview of a system that performs dynamic updating is presented. An updating algorithm and methods of ensuring program consistency throughout the update are discussed. An architecture to support the updating system is presented. A key concept of the architecture is the large, sparse virtual-address space. The functionality of this address space can be implemented with machine-independent virtual-memory primitives. The primitives provide a straightforward mapping from the updating concepts to existing virtual-memory hardware. The concepts and underlying architecture of the updating system are then exp and ed to work properly in a distributed computing environment. Since many programs that could benefit from an updating system are distributed programs, such scalability is essential. To validate these ideas, a prototype updating system was constructed and sample programs were updated with it. Sample sessions with the prototype are shown, and performance data is presented. The prototype demonstrated that it is indeed possible to dynamically update programs without imposing enormous constraints on the program, and without substantial performance degradation during the update.PhDComputer scienceUniversity of Michiganhttp://deepblue.lib.umich.edu/bitstream/2027.42/162358/1/9001706.pd