A research team from Sogang University's Computer Engineering Department has successfully had their paper accepted at the IEEE ICDE 2026 conference, focusing on storage technology advancements.
A research team led by Professor Park Seong-yong at Sogang University, particularly from the Data-centric AI Computing and Systems lab, has had their paper titled ‘Resystance: Unleashing Hidden Performance of Compaction in LSM-trees via eBPF’ accepted for presentation at the IEEE Conference on Data Engineering (ICDE) 2026. This achievement highlights the team's contributions to understanding and overcoming limitations in storage technologies, specifically concerning the inefficiencies in software that follows advances in hardware.
Storage technology has rapidly developed over recent years, particularly with the introduction of NVMe SSDs that offer high bandwidth and low latency, surpassing previous performance limitations. However, as hardware speeds have increased, the inefficiencies within software have become more pronounced. The research reveals that the bottleneck has shifted from hardware to the software stack, which includes the operating system, file systems, and system call interfaces, particularly in the context of LSM-tree based NoSQL databases like RocksDB.
The study investigates the structural bottleneck in LSM-tree based systems and finds that the compaction processes consume a significant portion of internal system calls within the LSM-tree. Specifically, the research identifies that the pread() system call, utilized while reading numerous data blocks, is a primary contributor to software stack bottlenecks. The findings suggest that the limitations in modern high-speed storage environments are not due to hardware, but rather reside between the operating system and databases, prompting a need to alleviate overhead within the software stack rather than pursuing new hardware solutions or modifying the LSM-tree structure.
