Some Thoughts on Distributed Recovery ( preliminary version )

This report deals with some aspects of distributed recovery. The report is divided into multiple parts, each part introducing a problem and a solution. The intent of this report is to present a medley of preliminary ideas, more detailed treatment may be presented elsewhere. The report deals with the following problems: A single processor failure tolerance scheme based on the distributed recovery unit abstraction. Staggered checkpointing and coordinated message logging to obtain a consistent logical recovery line. Relaxing the deenition of orphan messages. Exploiting network architecture to improve performance of message logging.