Luna-TX: An Optimized Transactional Mechanism for Persistent Memory

摘要

At present, most existing persistent memory programming libraries use write-ahead-logging(WAL) technology to ensure the consistency of memory allocating and updating process. In the application where persistent memory is frequently updated, this approach will bring a serious impact on system performance. In this work, we carefully analyzed the actual requirements of the applications and propose an optimized release mechanism for persistent memory called Luna_TX. In the improved mechanism, we remove the memory releasing phases of logs and objects from the critical path and generating in the process of update and release of transactions is delayed to the execution out of the transaction, which reduces the performance overhead on the critical paths. The experimental results have shown that the proposed mechanism can reduce the transaction delay significantly and boost up the whole transaction performance at 59%.