Recovery for Shared Disk Systems Using Multiple Redo Logs

A new method for redo logging and recovery is described. It is designed to work in a data sharing system where multiple nodes can access common data. Particularly important is that each nodecanhave its own log. Crash recovery is possible based on the contents of just one of these logs. Media recovery is supported via a form of merging of the logs. The method requires no time synchronization between nodes, and does not use timestamps to order the log records during the merge. The method should work with many undo recovery methods. Conventional checkpointing schemes can be adapted to the scheme, enabling the redo scan point to be advanced and the log to be truncated. Finally, the method makes possible a new paradigm for distributed DBMSs which has the potential to exploit inexpensive desktop processing and improve availability and responsiveness