Scheduling Heterogeneous Memory for Database Query Execution

To meet response time and throughput demands, data processing architects adapt query systems to new hardware. They have transitioned from disk-oriented to main memory-oriented architectures to leverage increasing memory capacities. Emerging technologies like high-bandwidth memory (HBM) and non-uniform  memory access (NUMA) enhance DRAM by balancing capacity, throughput, and  latency. However, this complexity poses challenges for database systems in fully utilizing the new hardware.

Hence, in this thesis we aim to tackle on of the challenges and try to  adapt an existing data processing engine for heterogeneous memory systems. The goal is to optimally distribute buffers, based on their access patterns, among available memory technologies.
This involves extracting buffer sizes and dependencies from the query execution plan to generate a memory schedule. The effectiveness of this schedule will be evaluated by adjusting the memory allocation in the query processing engine to place buffers in the  designated memories during execution