MIDAS: Minimizing Write Amplification in Log-Structured Systems through Adaptive Group Configuration

Explore a 21-minute conference talk from USENIX FAST '24 that introduces MIDAS, a novel approach to minimizing write amplification in log-structured systems. Learn how this innovative technique employs a chain-like structure of multiple groups to automatically segregate data blocks by age. Discover the analytical models, Update Interval Distribution (UID) and Markov-Chain-based Analytical Model (MCAM), used to dynamically adjust the number and size of groups based on workload I/O patterns. Understand how MIDAS isolates hot blocks into a dedicated HOT group, dynamically adjusting its size to minimize overall Write Amplification Factor (WAF). Examine the experimental results that demonstrate MIDAS outperforming state-of-the-art garbage collection techniques, offering 25% lower WAF and 54% higher throughput while consuming fewer resources. Gain insights into the potential impact of this research on improving the efficiency of log-structured systems in various applications, particularly in flash-based SSDs.