Benchmarking and Modeling Disk-based Storage Tiers for Practical Storage Design

摘要

This paper investigates benchmarking and modeling for a disk-based storage system in order to design and build a practical storage tier. As a practical case study, we focus on the design of an archival storage tier. The archival tiers play a critical role in data preservation as almost all current data will eventually be archived and the demands placed on archival tiers are growing because of large regularly-scheduled back-ups. Archival tiers usually consist of tape-based devices with a large storage capacity, but limited I/O performance for retrieving data, especially when multiple retrieval requests are made simultaneously. As the cost of disk-based devices continues to decrease while the capacity of individual disks increases, disk-based systems are becoming a more realistic option for both enterprise and commodity archival storage tiers. We utilize archival workloads developed from an analysis of historical data in order to provide accurate and robust benchmarks of system performance as an archive. We then embed our practical measurements in a measurement-driven optimization approach to design an archival system. Our approach produces a low cost design for a commodity disk-based archival storage system. Using our measurement-driven model, ideal storage building blocks are identified for a real-world archival tier design.