Browse Prior Art Database

Method for Avoiding False Notifications of Resource Exhaustion

IP.com Disclosure Number: IPCOM000019246D
Original Publication Date: 2003-Sep-08
Included in the Prior Art Database: 2003-Sep-08
Document File: 2 page(s) / 62K

Publishing Venue

IBM

Abstract

Software aging is the characteristic of some programs to become unreliable over time. Often this is due to unreleased resources, such as memory leaks. Actual resource utilization can be monitored and trend detection algorithms used to predict when resources are likely to be used up. When resource exhaustion is predicted, a system administrator or systems management software can be notified and corrective actions taken. Because resource utilization normally rises and falls even in healthy systems, it is possible that a false prediction of resource exhaustion can occur. In this case, unnecessary corrective actions may be taken. While normally these are less disruptive than an unexpected system failure, there is still a cost associated with these actions. With increasing use of automation to handle notifications, it is all the more important to minimize the occurance of false notifications. This paper describes a technique for reducing false notifications while also dynamically adjusting the algorithm's sensitivity with the level of resource utilization. This solution has been implemented in the IBM Director Software Rejuvenation tool.

This text was extracted from a PDF file.
At least one non-text object (such as an image or picture) has been suppressed.
This is the abbreviated version, containing approximately 52% of the total text.

Page 1 of 2

Method for Avoiding False Notifications of Resource Exhaustion

Avoiding False Notification of Resource Exhaustion Introduction

    Resource utilization in a computer system normally varies as applications are run, with increasing demands causing utilization to increase steadily. Software defects can cause similar results if unused resources are not released, a characteristic commonly referred to as software aging. One common example of software aging is memory leakage, in which unused memory is not released back to the operating system. Ultimately this leakage can lead to disruptive and costly system failures.

    Resource utilization can be monitored and trend detection algorithms used to predict when resources are likely to be totally used up. When resource exhaustion is predicted, a system administrator or management software can be notified and corrective actions taken. False Notification of Resource Exhaustion

    Because resource utilization rises and falls even in healthy systems, it is possible that a false prediction of resource exhaustion can occur, resulting in unnecessary corrective actions. While these are less disruptive than system failures, there is still a cost with these actions. Increasingly, notifications are being handled by systems management software. This makes it even more important that notifications are accurate, because automated handlers are often unable to filter out false notifications. It is highly desirable to minimize false notifications, while at the same time ensure that notifications are given if true resource exhaustion is likely to occur.

    The method described here is simple and efficient. When resource exhaustion is first predicted, a notify level is established. If actual resource usage reaches the notify level, a notification is sent. If resource usage remains below the notify level long enough that resource exhaustion is no longer predicted, the notify level is discarded and no notification is sent. The sensitivity of this notify level is dynamic. Because it is set relative to the current resource usage, it is more sensitive when there are fewer remaining resources and less sensitive when there are more remaining resources. This approach achieves the desired goal of advanced notification with reduced false notifications.

Description

    In the solution, a small program periodically collects data on the target system and predicts when a resource will exhaust. A notify level is set at the amount of resource utilization that triggers notification.

    In Figure 1 below, the upward sloping line to the left of Tc represents actual resource utilization. The upward sloping line to the right of Tc represents predicted resource utilization.

The Algorithm
1. In the figure, resource utilization data is extrapolated to Th, the prediction horizon. This...