A Hash-Based Clustering System Software for Intermittent Computing Devices With NAND Flash Memory
Chin-Hsien Wu,Chia-Cheng Liu,Po-Cheng Yu
DOI: https://doi.org/10.1109/tcad.2024.3380553
IF: 2.9
2024-01-01
IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems
Abstract:In recent years, the intermittent computing devices have become increasingly widespread and result in a greater demand for storage space. In particular, NAND flash memory is well-suited as a storage medium for the intermittent computing devices, but its "out-of-place" update characteristic leads to a garbage collection (GC) mechanism to reclaim invalid pages when the number of free pages becomes insufficient. The previous studies have shown that a suitable separation of hot and cold data (i.e., a clustering method) can significantly reduce the overhead of garbage collection and improve the performance. However, the previous studies are not suitable for the intermittent computing devices that are equipped with the limited volatile memory space and very low computing power. Therefore, we will propose a hash-based clustering system software for the intermittent computing devices to provide a suitable separation of hot and cold data. The experimental results show that the proposed method with low computational time and low volatile memory space can achieve a reduction of 8% 10%, 8% 10%, 10%, and 5% 7%, in page reads, page writes, block erases, average number of erases per block, and a reduction of 8% 11% in write amplification when compared to the previous methods (such as a DBSCAN-based clustering method and a K-means clustering method). Additionally, the proposed method can efficiently recover to the most recent state when the intermittent computing devices experience power outages.
engineering, electrical & electronic,computer science, interdisciplinary applications, hardware & architecture