Reliability mechanisms for ADAMS

The goal of checkpointing in database management systems is to save database states on a separate secure device so that the database can be recovered when errors and failures occur. This paper presents a non-interfering checkpointing mechanism being developed for ADAMS. Instead of waiting for a consistent state to occur, our checkpointing approach constructs a state that would result by completing the transactions that are in progress when the global checkpoint begins. The checkpointing algorithm is executed concurrently with transaction activity while constructing a transaction-consistent checkpoint on disk, without requiring the database quiesce. This property of non-interference is highly desirable to real-time applications, where restricting transaction activity during the checkpointing operation is in many cases not feasible. Two main properties of this checkpointing algorithm are global consistency and reduced interference, both of which are crucial for achieving high availability.