Abstract: An approach for consistent database recovery for distributed database systems uses “synchronization points”. A synchronization point is a global timestamp for which across all nodes of a distributed database system, the nodes have stored change records for any transaction occurring at and before the synchronization point in persistent logs. Each node may employ asynchronous flushing. However, on a periodic basis, each node coordinates to establish a synchronization point, which may entail ensuring change records for transactions that committed at or before the synchronization point are stored in persistent logs. In effect, a synchronization point represents that any transaction committed at or before the synchronization point has been durably committed.

Abstract: Under block-aware encoding, a bitmap represented by atoms comprises a series of bitmaps for each data block in a database. Each bitmap in the series is referred to herein as a block bitmap. Each block bitmap may have a different number of bytes or bits. Gaps are represented in atoms using a pair of numbers referred to as a gap code. A gap code includes a block-skip code and slot-skip code. A block-skip code represents how many block bitmaps to advance to reach a subsequent block bitmap; a slot-skip code represents how many bytes to advance within the block bitmap to reach a byte with at least one bit set. A gap code is represented by bit positions within a byte, with some bit positions allocated to represent the block-skip code and some to represent the slot-skip code. The allocation is adjusted dynamically during encoding and decoding.