| As memory is getting cheaper and larger, it is the trend that memory gradually replaces disk as the primary storage for database systems. Logging for the durability and concurrency control for the runtime efficiency are the two components that restrict the performance of in-memory databases. In this thesis, first propose Adaptive Logging techniques to achieve an optimized trade-off between runtime and recovery performance. Then we proposed a novel Dependency Graph Based Concurrency Control (DGCC) protocol for multi-core in-memory databases. With the experience working on logging and concurrency control techniques, we found that recovery is dependent on the sequence in which transactions are executed. Hence, we propose to treat the transaction management and recovery problems as one by introducing distributed DGCC and Dependency Logging. |
