Browse Prior Art Database

Real Memory Contention Control

IP.com Disclosure Number: IPCOM000081761D
Original Publication Date: 1974-Aug-01
Included in the Prior Art Database: 2005-Feb-28
Document File: 2 page(s) / 14K

Publishing Venue

IBM

Related People

Bowen, AJ: AUTHOR

Abstract

This is a control mechanism in a multiprogramming computer system with virtual memory for managing a group of batch tasks, in such a way that system resource utilization is optimized under conditions of real memory contention.

This text was extracted from a PDF file.
This is the abbreviated version, containing approximately 52% of the total text.

Page 1 of 2

Real Memory Contention Control

This is a control mechanism in a multiprogramming computer system with virtual memory for managing a group of batch tasks, in such a way that system resource utilization is optimized under conditions of real memory contention.

In a multiprogrammed computer, if memory contention occurs, it often manifests itself as an overload on the paging device. Under this condition, many operating systems simply render one or more tasks dormant, i.e., the selected tasks are temporarily rendered ineligible for service by any system resource. While this relieves the problem, such deactivation is unduly severe and in many instances explicit control of the real memory resource assigned to active tasks is all that is required.

The mechanism for optimizing system resource utilization under conditions of real memory contention is to provide two levels of control. The first level, termed the Selector level, provides for coarse adjustment of the operating point, a term used to express the activity level of the system. It has three primary dimensions: CPU utilization, I/O subsystem utilization, and memory/paging subsystem utilization.

The Selector level is invoked if there is insufficient real memory available, in which case deactivation of a task is required; or it is also invoked if there is excess real memory available, in which case reactivation of a task is required.

CPU utilization and nonpaging I/O subsystem utilization are the inputs to the mechanism. The mechanism operates by controlling which tasks belong to the currently active set at a given moment, and attempts to shift the active task mix so that CPU resource and nonpaging l/O resource loads are maximized. The outputs of this level of control determines (1) whether no action is to be taken; (2) whether a task is to be deactivated; (3) whether a task is to be reactivated; and
(4). provides statistical data in the event of memory/paging subsystem throttling.

The second level of control, termed the Fine Tuning level, provides an ability to fine tune the operating point around the coarse point arrived at via the first Selector level of control. This level of control is invoked if a page fault, job step or phase start, job step or phase end, or timer interrupt has occurred. The mechanism monitors a moving average vector (MAV) of task CPU time between page faults, in order to control changes in real memo...