Browse Prior Art Database

Learning Model for Monitoring System Health

IP.com Disclosure Number: IPCOM000238732D
Publication Date: 2014-Sep-15
Document File: 4 page(s) / 110K

Publishing Venue

The IP.com Prior Art Database

Related People

Chethan Suresh: INVENTOR

Abstract

A learning model is disclosed for monitoring system health by learning about state of a system and predicting state of the system. The learning model is configured to estimate time of failure of the system and estimate a threshold for metrics that are used to monitor health of the system.

This text was extracted from a Microsoft Word document.
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.

Learning Model for Monitoring System Health

Abstract

A learning model is disclosed for monitoring system health by learning about state of a system and predicting state of the system.  The learning model is configured to estimate time of failure of the system and estimate a threshold for metrics that are used to monitor health of the system. 

Description

Disclosed is a learning model for monitoring system health by learning about state of a system and predicting state of the system.  The system may be one of, but not limited to, Operating System (OS), domain specific applications or any system process that provides a range of services.  The learning model is configured to estimate time of failure of the system and estimate a threshold of metrics that are used to monitor health of the system.  The learning model is created by building a knowledge base that includes state of similar/homogenous systems defined by a feature vector as illustrated in Fig. 1. 

Figure 1

As illustrated in Fig. 1, determinative features of a system are identified and the feature vector collects data regarding the features.  Each feature vector is a representation of system/process's current state from which TFF is estimated.  The TFF estimate is done at each stage using previous knowledge or experience gained to predict such system’s time for failure.  The feature vectors that are built are sent to monitoring dashboards and shared/distributed storage systems. 

These feature vectors are collected periodically and stored.  Thereafter, a vector space model of these feature vectors is built and a k-Nearest Neighbor (k-NN) model is built to estimate time to failure based on the knowledge base. 

In k-NN model, the output is a class membership.  Further, an object is classified by a majority vote of its neighbors, with the object being assigned to a class most common among its k nearest neighbors.  If k = 1, then the object is simply assigned to a class of single nearest neighbor.  The output is the property value for the object.  This value is an average of the values of its k nearest neighbors.

Results obtained from the vector space model and the k-NN model is evaluated to check for anomalies.  Upon checking for anomalies, new feature vectors may be built based on the results.  These feature vectors may be adapted to improve future estimates of TFF values.  This may be achieved by using a multi-class classifier such as Naïve Bayes or Decision tree model of classifier.  The class label may be a range of TFF values for a giv...