Data Management Seminar: Kaisong Huang, Systematic Latency Optimizations for Modern Database Systems
Content
Speaker
Kaisong Huang
Title
Systematic Latency Optimizations for Modern Database Systems
Abstract
Modern database systems are the backbone of countless applications, from financial transactions to e-commerce and real-time analytics. As workloads grow more diverse, meeting the demands for high-throughput, low-latency performance becomes increasingly challenging. Key obstacles arise from the need to handle complex interactions between compute, memory, and storage resources efficiently, especially in mixed workloads with varying transaction profiles and priorities.
In this seminar, I will present my research on optimizing latency for modern database systems built on multicore CPUs, large memory, and NVMe SSDs. I will begin by identifying the major sources of latency in this context, focusing on data access and CPU scheduling latencies. To address data access latency, I introduce MosaicDB, which utilizes lightweight cooperative scheduling mechanisms to interleave data loading and computation in a memory-optimized online transaction processing engine. To further reduce CPU scheduling latency for workloads with high-priority short transactions and low-priority long-running transactions, I propose PreemptDB, the first database system to enable lightweight preemptive scheduling using hardware-based interrupts on recent Intel x86 CPUs. Finally, I will outline my future research directions aimed at building fast, scalable data-intensive systems that leverage new computing architectures and hardware technologies.
Bio
Kaisong Huang is a PhD candidate at Simon Fraser University supervised by Prof. Tianzheng Wang. Kaisong's research focuses on data-intensive systems, specifically in database engines, transaction processing, storage management, and disaggregated systems. His work has appeared in leading database venues including CIDR, SIGMOD and VLDB. He received ACM SIGMOD Research Highlight Award in 2023. More information about his research can be found at https://kaisonghuang.github.io.
Host