Research And Implementation Of Alarm Integration Technique For Cluster Based Association Analysis

With the development of informationization in every society industry, cluster system with high-performance and high avalability has been becoming the improtant underlayed platform of informationization. However, the architecture of cluster is becoming more and more complicated, and the quantity of hardware and software is expanding. As a result, these cases bring along a series of problems, such as the rising of system fault rate and alarm quantity, the difficulty of fault location. Further more, these problems bring about much difficulty to management for cluster system.To deal with these problems, the thesis researches servral key technologies about the unified form conversion of alarms, alarm integration, association analysis and so on. What is more, an Alarm Information Processing System for Cluster (CAIPS) is offered. This system is used to integrate the cluster alarms, and not only decreases the alarm quantity, but also increases the semantic of alarm. All of these do great favor to system manager for diagnosing, locating and eliminating faults. The concrete works are as follows:1. The thesis brings forward a model for dealing with cluster alarm. Furthermore, architecture of alarm information proceesing system for cluster is designed based on this model.2. By leading the concepts of time restrict and support number into the GSP (Generalized Sequential Patterns) algorithm, the thesis puts forward a Frequent Pattern Mining Algorithm based on Single Alarm Sequence. This algorithm could mine the association rules between the cluster alarms effectively, and it is proved to be effective by corresponding test.3. To deal with heterogeneous alarm, the thesis offers a kind of unified cluster alarm information strcuture and a kind of form for the main alarm elements. With strong adaptability and expansibility, the structure could describe the most cluster alarm.4. The thesis introduces a mechanism for integrating alarms based on rules, and researches the types of redundancy and correlative alarms. The mechanism is proved to be effective in filtrating and reconstruct the redundancy and correlative alarms, and it satisfies the requirment of fault location.We implement the CAIPS system, which is applied to the MDMP (Massive Data Management Platform) successfully, and becomes one part of the MDMP management system.