Revisiting Page Migration for Main-Memory Database Systems

Modern NUMA servers and tiered memory systems have significantly improved the DBMS performance, and are poised to deliver further improvements in the future. However, realizing this potential depends on the DBMS's ability to efficiently migrate pages among different NUMA nodes, and/or memory chips as the workload evolves. Modern DBMSs blindly offload the migration procedure to the operating system without accounting for the workload and its migration characteristics. In this paper, we propose a custom system call move_pages2 as an alternate to Linux's own move_pages system call. Experiments on a B$^+$-Tree for a YCSB-like workload show that the proposed move\_pages2 custom system call outperforms the native Linux system call by up to 1.84$\times$ and 2.09$\times$, in terms of query and page migration throughputs, respectively.