Performance isolation: sharing and isolation in shared-memory multiprocessors

  • Authors:

  • Ben Verghese;Anoop Gupta;Mendel Rosenblum

  • Affiliations:

  • Western Research Laboratory, Compaq Computer Corporation, Palo Alto, CA;Computer Systems Laboratory, Stanford University, Stanford, CA and Microsoft Corporation, Redmond, WA;Computer Systems Laboratory, Stanford University, Stanford, CA

  • Venue:
  • Proceedings of the eighth international conference on Architectural support for programming languages and operating systems
  • Year:
  • 1998

Quantified Score

Hi-index 0.00

Visualization

Abstract

Shared-memory multiprocessors (SMPs) are being extensively used as general-purpose servers. The tight coupling of multiple processors, memory, and I/O provides enormous computing power in a single system, and enables the efficient sharing of these resources.The operating systems for these machines (UNIX or Windows NT) provide very few controls for sharing the resources of the system among the active tasks or users. This unconstrained sharing model is a serious limitation for a server because the load placed by one user can adversely affect other users' performance in an unpredictable manner. We show that this lack of isolation is caused by the resource allocation scheme (or lack thereof) carried over from singleuser workstations. Multi-user multiprocessor systems require more sophisticated resource management, and we show how the proposed "performance isolation" scheme can address the current weaknesses of these systems. We have implemented performance isolation in the Silicon Graphics IRIX operating system for three important system resources: CPU time, memory, and disk bandwidth. Running a number of workloads we show that our proposed scheme is successful at providing workstation-like isolation under heavy load, SMP-like latency under light load, and SMP-like throughput in all cases.