Loading...

Enhancing Performance of Replacement Policy in Operating System Page Cache for I/O Intensive Workloads

Salehi Mazdeh, Zahra | 2016

587 Viewed
  1. Type of Document: M.Sc. Thesis
  2. Language: Farsi
  3. Document No: 48840 (19)
  4. University: Sharif University of Technology
  5. Department: Computer Engineering
  6. Advisor(s): Asadi, Hossein
  7. Abstract:
  8. Due to the increasing use of data storage systems and volume of digital data, fast information access and retrieval has significant effect on performance of storage-bound systems. Operating systems utilize the spare space in the main memory as a cache to speed up access to the storage subsystem. Due to the large performance gap between hard disk and main memory, any improvement in this caching layer can greatly affect the performance of the entire system. Page replacement algorithms are one of the important mechanisms in the virtual memory management layer. Several algorithms have been proposed for the eviction of pages from the virtual memory cache and few of them have also been implemented in the operating system and database systems. However, due to limitations such as lack of proper characterization of virtual memory workloads, design, evaluation, and optimization of these algorithms is challenging. In this thesis, we perform an extensive study of virtual memory subsystem in Linux as a modern and performant operating system. We then use information extracted from the Kernel code and documentation to simulate the behavior of the operating system page cache. By observing the behavior of workloads in virtual memory management layer and calculating properties such as page access frequency and re-use distance, we analyze the behavior of IO intensive workloads. To benchmark the performance of the Linux algorithm and our proposed algorithm, we implement the Clock-Pro and Belady algorithms and use the hit ratio to evaluate the performance of algorithms. The results show that our proposed algorithm improves cache hit ratio up to 10%
  9. Keywords:
  10. Characterization ; Fast Retrieval ; Storage Systems ; Virtual Memory Management ; Page Replacement Algorithm ; Input/Output Intensive Workloads

 Digital Object List

 Bookmark

No TOC