Towards Scalable Manycore-aware Persistent B plus -Trees for Efficient Indexing in Cloud Environments

초록

The emergence of manycore machines with Intel DC Persistent Memory (DCPM) aims to provide high performance and scalability with persistence guarantees. Thus, it is required to offer opportunities to port DRAM-based index data structures to DCPM to fully exploit the performance of these machines. Fast & Fair (F&F) is the state-of-the-art concurrent variant of the B+-tree for DCPM. However, its adoption on manycore machines suffers from scalability limitations due to lengthy, lock-based synchronization including structure modification operations (SMOs). In this work, we propose F-3-tree, a concurrent, persistent future-based B+-tree that shows superior scalability on DCPMs. F-3-tree design relies on thread-local future objects and a global B+-tree. We employ an in-memory hash table to mitigate the read overhead for the key searches in thread-local future objects. We implemented the proposed ideas atop F&F and performed experiments on Linux (kernel v5.4.0) using both synthetic and real-world workloads. We evaluated F-3-tree with F&F and the results show that F-3-tree outperforms F&F by 3.4x on average for sequential, random, and mixed workloads.

키워드