Hot and Cold Data Migration Strategy Based on Newton’s Law of Cooling
Shuangjian Wei, Hongpeng Lin, Aidi Tan · 2024
We present a novel temperature-based data migration strategy for tiered storage systems that addresses the limitations of existing Newton’s Law of Cooling approaches. While current methods treat temperature increase from data access as a constant, our approach introduces a logistic function-based model that dynamically adjusts temperature increases based on access patterns. This enhancement prevents artificial temperature inflation during burst accesses and better reflects real-world data access behaviors. Experimental results show our Temperature Migration Strategy (TMS) achieves superior identification effectiveness compared to LRU-2 and ARC across various configurations, with particular resilience to sudden access pattern changes. The proposed solution demonstrates superior stability under extreme workloads, with only a 3.4% average performance degradation compared to 31.2% in traditional approaches. Our work provides a robust foundation for efficient hot/cold data management in modern storage hierarchies, particularly beneficial for OLTP systems with skewed access patterns.