2026-05-24T08:47:26Zhttps://keep.lib.asu.edu/oai/request

oai:keep.lib.asu.edu:node-2030622025-12-15T17:22:25Zoai_pmh:alloai_pmh:repo_items

203062 https://hdl.handle.net/2286/R.2.N.203062 http://rightsstatements.org/vocab/InC/1.0/ All Rights Reserved 2025 75 pages Masters Thesis Academic theses en Sutila, Alexander Zhao, Ming Cao, Zhichao Zou, Jia Arizona State University Partial requirement for: M.S., Arizona State University, 2025 Field of study: Computer Science Flash caching is widely used to accelerate data accesses to remote storage by caching data on local solid state drives (SSDs). However, caching worsens the endurance problem that SSDs face because cache replacement causes substantial writes and write amplification (WA). The emerging Zoned Namespace (ZNS) SSDs promise new opportunities for optimizing SSD caching as they expose sequentially writable zones for hosts to place data and manage garbage collection (GC). Conventional GC policies are not optimized for SSDs used for caching; existing SSD caching solutions cannot adequately exploit the locality characteristics of workloads to minimize WA without compromising performance. New research is needed to optimize caching policies for ZNS SSDs which are not widely available. Moreover, running experiments to test new caching policies on hardware can reduce device lifetimes even further, incur lengthy test durations, and require lots of time and effort to produce deployable prototypes. To alleviate the burdens of testing on real hardware, the performance of caching policies is often evaluated via simulation. However, existing simulation software does not provide insight into WA or GC efficiency and is therefore insufficient for exploring the design space made available by ZNS-SSDs. This thesis proposes ZNS-Cache-Sim (ZNSCS), a modular caching simulation framework for ZNS-SSD-based caching solutions. ZNSCS models data placement and GC to evaluate cache performance and device endurance characteristics of new caching policies. It offers easily extendable and hot-swappable cache components and supports replaying real-world request traces and synthetically generated workloads. This thesis also presents various new algorithms that can be used in ZNS-SSD-based caching solutions designed to improve WA without degrading overall cache performance. ZNSCS is used to explore the available design space and evaluate each proposed algorithm to show their ability to balance performance with endurance and justify their design. Computer Science Algorithms Caching Optimization Simulation SSD Storage A Versatile ZNS SSD Based Cache Simulation Framework