A New Readahead Framework for SSD-based Caching Storage in IoT Systems
Preprint
- 23 January 2024
- preprint
- Published by MDPI AG in Preprints.org
Abstract
In an IoT system, the sheer volume of data generated by numerous sensing devices necessitates a well-designed scheme for storing and retrieving data efficiently, enabling streamlined data processing and analytics. One promising storage architecture involves utilizing solid-state drives (SSDs) to cache data from hard disk drives (HDDs), thereby creating an SSD-based caching storage system. To further enhance access performance, it is possible to employ readahead techniques to minimize data access latency. However, the existing Linux readahead scheme falls short in fully leveraging SSD-based caching storage systems. We address this limitation by introducing a novel cross-layered readahead architecture that effectively communicates with the VFS layer, the file system layer, and the block I/O layer. This communication facilitates the acquisition of readahead timing, readahead data continuity, and readahead data location, respectively. To guide prefetching decisions, our architecture analyzes the degree of data access sequentiality, the performance model of the target storage device, and the access patterns of the I/O workload on the corresponding storage device. The implementation of this new architecture in the Linux kernel yields promising experimental results, demonstrating its robustness by consistently outperforming the stock Linux kernel. Notably, our architecture reduces the total execution time of the stock Linux kernel by up to 49%, except in cases of random workloads where both the stock Linux kernel and our architecture exhibit similar performance.Keywords
All Related Versions
This publication has 28 references indexed in Scilit:
- A Prefetching Scheme for Multi-Tiered Storage Systems2018
- PARC: A novel OS cache managerSoftware: Practice and Experience, 2018
- Towards a fast and secure design for enterprise‐oriented cloud storage systemsConcurrency and Computation: Practice and Experience, 2017
- A flexible I/O arbitration framework for netCDF‐based big data processing workflows on high‐end supercomputersConcurrency and Computation: Practice and Experience, 2017
- Lynx: a learning linux prefetching mechanism for SSD performance model2016
- Server‐side prefetching in distributed file systemsConcurrency and Computation: Practice and Experience, 2014
- Profit data caching and hybrid disk-aware Completely Fair Queuing scheduling algorithms for hybrid disksSoftware: Practice and Experience, 2014
- Flashy prefetching for high-performance flash drives2012
- Integrating flash-based SSDs into the storage stack2012
- STEP: Sequentiality and Thrashing Detection Based Prefetching to Improve Performance of Networked Storage Servers2007