Adaptive Data Placement Techniques for Memory-Scalable Big Data Management

The increasing demands of big data applications necessitate innovative memory management solutions that combine performance, scalability, and cost efficiency. Traditional DRAM-only systems face challenges in cost and power consumption, while emerging Non-Volatile Memory (NVM) offers durability and scalability. This paper presents an Adaptive Data Placement Framework designed for hybrid DRAM-NVM systems. By leveraging real-time monitoring, predictive analytics, and machine learning, the framework dynamically classifies and allocates data to optimize memory utilization. Experimental results demonstrate significant performance improvements, including 23% reduced latency, 16% improved energy efficiency, and 31% lower write amplification compared to state-of-the-art systems like HeMem and HM-Keeper. These advancements highlight the framework’s potential to address the complex demands of modern big data applications, bridging the gap between scalability and efficiency in hybrid memory systems.