Enhanced process-aware memory deduplication with dynamic fine-tuning method: implementation and applications
Yuquan Dong, Guangyin Shi, Xiao Xue, Chuanji Gao, Weiwei Cai, Yongkang Hou, J Wang · 2025
Modern cloud computing environments demand efficient memory resource utilization and cost-effectiveness. Existing memory merging technologies, such as Kernel Samepage Merging (KSM), suffer from limitations in flexibility and dynamic adaptation to diverse applications. This paper introduces Process-Aware Memory Deduplication (PAMD), a novel approach that employs dynamic fine-tuning to address these challenges. PAMD implements a memory merging service architecture that integrates a new kernel module, enabling specific running processes to dynamically join or exit the KSM scanning list, thus achieving more granular memory merging management. Furthermore, PAMD incorporates a dynamic adjustment algorithm that prioritizes memory merging benefits for individual processes. By setting configurable threshold parameters, the algorithm dynamically adjusts the memory merging status of processes, improving efficiency. Experiment results demonstrate that PAMD significantly surpasses traditional KSM methods in terms of flexibility and applicability of memory merging. PAMD technology achieves more efficient memory management and higher memory utilization in modern cloud computing environments.