Loading...

Improving Efficiency of Storage Subsystems for Virtualized Platforms

Ahmadian Khameneh, Saba | 2019

509 Viewed
  1. Type of Document: Ph.D. Dissertation
  2. Language: Farsi
  3. Document No: 52950 (19)
  4. University: Sharif University of Technology
  5. Department: Computer Engineering
  6. Advisor(s): Asadi, Hossein
  7. Abstract:
  8. In recent years, high interest in using virtualization platforms and Virtual Machines (VMs) in enterprise data centers and Clouds has significantly increased the demand for high performance data storage systems. Recent studies in virtualized platforms have suggested using high-performance storage elements such as Solid-State Drives (SSDs) as a caching layer for storage subsystems mainly built upon Hard Disk Drives (HDDs). Such studies implicitly focus on sharing and partitioning the I/O cache space between VMs, while they neglect to address other important parameters such as SSD lifetime, reliability, and cache policies, which can significantly affect the efficiency of I/O caching in terms of performance and cost. In this thesis, we propose efficient I/O caching architectures to manage performance, reliability, and endurance of the storage subsystem in virtualized platforms. Our proposal by using high-performance storage devices such as SSDs and Dynamic Random Access Memories (DRAM) in the caching layer and by considering the characteristics of the workloads 1) effectively improves performance and endurance and 2) assigns the most efficient cache size and policy for the running VMs. Furthermore, to evaluate the reliability of the I/O caching layer and examine the impact of workload I/O characteristics on the reliability, we propose a physical test platform, which injects two most common data center failures, sudden power outage and high operating temperature faults, into the SSDs and detects different types of failures in the caching layer. Our experiments on real systems reveal that our proposed I/O caching architecture for virtualized platforms significantly improves performance (by 45%), cost (by 51%), and endurance (by 33%), compared to the state-of-the-art caching schemes
  9. Keywords:
  10. Performance ; Endurance ; Reliability ; Storage Systems ; Virtualization ; Input/Output Cache ; Virtualized Platforms

 Digital Object List

 Bookmark

...see more