Improving database performance and space efficiency in deduplicated and snapshotable memory systems
- In-memory databases have become increasingly important and prevalent because of the reduced cost of DRAM and the increasing demand for performance. Different from disk-oriented databases, in-memory databases treat main memory as the major storage for most or all of its data. This new storage organization faces several challenges, including DRAM scaling slowdown, concurrency control bottleneck, and index performance dilemma. To address these challenges, this thesis resorts to the recently proposed memory architecture called HICAMP. The HICAMP memory provides fine-grain data duplication, lightweight memory snapshots, and DAG structures in hardware. These features enable new ways to design and implement in-memory databases. This thesis introduces HicampDB, an in-memory database built from scratch to leverage HICAMP's architecture features. HicampDB organizes its storage and index on the HICAMP DAGs to benefit from the efficient tree traversal and data deduplication. HicampDB's bitmap index supports online updates while preserving high compaction ratio. Thanks to the hardware memory snapshots, HicampDB transactions run in full isolation and require no locking or database partitioning. The hybrid conflict detection used in HicampDB avoids the scalability bottleneck of OCC systems. In the experiments, HicampDB demonstrates 10x higher space efficiency, 16x better access speed, and 70% lower memory bandwidth consumption for database index. On the TPC-C benchmark, HicampDB shows near-linear scaling to 32 cores and 4.8x higher throughput when comparing with a state-of-the-art in-memory database.
|Type of resource
|electronic; electronic resource; remote
|1 online resource.
|Wang, Bo, (Computer systems engineer)
|Cheriton, David R
|Cheriton, David R
|Stanford University, Department of Electrical Engineering.
|Statement of responsibility
|Submitted to the Department of Electrical Engineering.
|Thesis (Ph.D.)--Stanford University, 2016.
- © 2016 by Bo Wang
- This work is licensed under a Creative Commons Attribution Non Commercial 3.0 Unported license (CC BY-NC).
Also listed in
Loading usage metrics...