Improving I/O performance using virtual disk introspection

  • Authors:
  • Vasily Tarasov;Deepak Jain;Dean Hildebrand;Renu Tewari;Geoff Kuenning;Erez Zadok

  • Affiliations:
  • Stony Brook University;Stony Brook University;IBM Research-Almaden;IBM Research-Almaden;Harvey Mudd College;Stony Brook University

  • Venue:
  • HotStorage'13 Proceedings of the 5th USENIX conference on Hot Topics in Storage and File Systems
  • Year:
  • 2013

Quantified Score

Hi-index 0.00

Visualization

Abstract

Storage consolidation due to server virtualization puts stringent new requirements on Storage Array (SA) performance. Virtualized workloads require new performance optimizations that cannot be totally addressed by merely using expensive hardware such as SSDs. This position paper presents Virtual Machine Disk Image (VMDI) introspection--a key technique for implementing a variety of virtualization-oriented I/O optimizations. VMDI introspection gives SAs an understanding of guest file system semantics, such as determining object types and classifying read and write operations. We explore possible approaches for VMDI introspection and then describe a set of VMDI-introspection-based optimizations. Our prototype implementation with enhanced meta-data caching and placement shows 11% to 20× performance improvement.