Is the optimism in optimistic concurrency warranted?

  • Authors:
  • Donald E. Porter;Owen S. Hofmann;Emmett Witchel

  • Affiliations:
  • Department of Computer Sciences, The University of Texas at Austin, Austin, TX;Department of Computer Sciences, The University of Texas at Austin, Austin, TX;Department of Computer Sciences, The University of Texas at Austin, Austin, TX

  • Venue:
  • HOTOS'07 Proceedings of the 11th USENIX workshop on Hot topics in operating systems
  • Year:
  • 2007

Quantified Score

Hi-index 0.00

Visualization

Abstract

Optimistic synchronization allows concurrent execution of critical sections while performing dynamic conflict detection and recovery. Optimistic synchronization will increase performance only if critical regions are data independent--concurrent critical sections access disjoint data most of the time. Optimistic synchronization primitives, such as transactional memory, will improve the performance of complex systems like an operating system kernel only if the kernel's critical regions have reasonably high rates of data independence. This paper introduces a novel method and a tool called syncchar for exploring the potential benefit of optimistic synchronization by measuring data independence of potentially concurrent critical sections. Experimental data indicate that the Linux kernel has enough data independent critical sections to benefit from optimistic concurrency on smaller multiprocessors. Achieving further scalability will require data structure reorganization to increase data independence.