Proactive leader election in asynchronous shared memory systems
ATVA'06 Proceedings of the 4th international conference on Automated Technology for Verification and Analysis
| Hi-index | 0.00 |
The authors classify object failures into two broad categories: responsive and non-responsive. They require that wait-free objects subject to responsive failures continue to respond (in finite time) to operation invocations. The responses may be incorrect. In contrast, wait-free objects subject to non-responsive failures are exempt from responding to operation invocations. Such objects may 'hang' on the invoking process. They divide responsive failures into three models: R-crash,R-omission, and R-arbitrary. They divide non-responsive failures into crash, omission, and arbitrary. An object subject to crash failure behaves correctly until it fails, and once it fails, it never responds to operation invocations. An object subject to omission failures may fail to respond to the invocations of an arbitrary subset of processes, but continue to respond to the invocations of the remaining processes (forever).