PDOS: Experimenting with Garbage Collection in an Otherwise Conventional OS

University Of Utah, Salt Lake City UT

Investigators

Abstract

Security holes and crashes at the operating system (OS) level represent a serious infrastructure problem. These OS bugs are often the result of memory management errors, which are very hard for developers to avoid. In the long run, a broad class of memory-management errors can be prevented by using a high-level, type-safe language, but whether these languages are suitable for kernel implementation remains an open question. Unfortunately, any given experiment to answer the question (i.e., using a particular high-level language) involves many differences in implementation compared to a conventional kernel, making it difficult to draw conclusions about which parts of the high-level approach work and which parts do not. This project is an experiment specifically about garbage collection for legacy OS kernels, changing as few other implementation issues as possible. Concretely, the research is about developing tools to automatically transform the C source code of a conventional kernel so that it is compatible with a variety of precise garbage-collection strategies. The project's ultimate goal is to check whether the OS becomes usefully more reliable as a result of garbage collection, and to measure the performance costs in both desktop and embedded environments. If the experiment shows that the OS improves at a reasonable cost, then the research will have shown how to increase the reliability of a major part of today's computer systems infrastructure. Tools generated by the project will be made publicly available for use in research, for application to practical software, and for pedagogical purposes.