Date of Award
Thesis - SCU Access Only
Santa Clara : Santa Clara University, 2018.
Master of Science (MS)
This study explores the use of sequential writes in hybrid storage consisting of HDD bulk storage and a low-end SSD with poor random write performance acting as a cache. The present work modifies an existing algorithm, DCD (Disk Caching Disk.) DCD makes low-end SSDs, which have poor random write performance, act as a cache as though they were high-end SSDs, which have good random write performance. DCD uses sequential writes to cache data, relying on the good sequential write performance of SSDs. This work describes the design and implementation of a novel caching algorithm that keeps the internal sequentiality of individual log files from DCD, but does so while fragmenting the placement of log files in the cache for the sake of a reserved section that protects frequently accessed data from being evicted. The size of this reserved section is then adjusted adaptively based on the workload. This adaptive scheme is borrowed from the algorithm ARC (adaptive replacement cache.) Our new algorithm, called AR-DCD (Adaptive Replacement Disk Caching Disk,) is measured to have a hit rate that is competitive with that of DCD. For most workloads, AR-DCD has a hit rate superior to that of DCD. However, the speed of AR-DCD is slower than that of DCD due to AR-DCD’s scattered access pattern on the cache device. Hence, AR-DCD cannot be recommended for use in production systems. Nevertheless, the superior hit rate demonstrates the fundamental soundness of the adaptive caching scheme inspired by ARC, and future work may be able to improve on AR-DCD to have better performance than DCD.
Han, Allen, "The Design and Implementation of an Adaptive Log Cache" (2018). Computer Science and Engineering Master's Theses. 8.
SCU Access Only
To access this paper, please log into or create an account in Scholar Commons using your scu.edu email address.