Clustered objects

  • Authors:
  • Jonathan Appavoo

  • Affiliations:
  • University of Toronto (Canada)

  • Venue:
  • Clustered objects
  • Year:
  • 2005

Quantified Score

Hi-index 0.00

Visualization

Abstract

In this dissertation we establish that the use of distribution in the implementation of a shared memory multi-processor operating system is both feasible and able to substantially improve performance of core operating system services. Specifically we apply distribution in the form of replication and partitioning in the construction of K42, a shared memory multi-processor operating system. Clustered Objects, a software construction for the systematic and selective application of distribution to objects of K42's object oriented system layer, is presented. A study of the virtual memory services of K42 is conducted in which distribution is applied to key virtual memory objects to enable performance improvements. The distributed versions of these objects substantially improve scalability, and specifically improve throughput of a standard multiuser benchmark by 68% on a 24 way multi-processor. Additionally, a methodology for the dynamic hot-swapping of Clustered Object instances is presented as a means for enabling dynamic adaptation. Motivated by the desire to hot-swap between centralized and distributed implementations of Clustered Objects, the methodology presented is correct, efficient and integrated with the Clustered Object support.