TUM Logo

Concept Learner for CorMeL Transaction Trees in Amadeus Data Processing Systems

The architecture of Amadeus Data Processing Systems is distributed and highly coupled. To process an incoming request, various systems communicate with each other and process this request. CorMeL is a system which correlates all the logs gathered between those system components processing that incoming request. A request might occasionally not get processed due to changed configuration, hardware or software failures, and so on, and it is called as an Incident. In this thesis, approaches for anomaly detection in distributed information systems are analyzed and developed. This thesis also seeks to develop information retrieval methods from anomalous labeled data to locate possible failure causes for an incident in these systems. The results show that even though effectiveness of methods to locate failure causes is quite substantial, however, for certain scenarios, it demands deeper research. Discussion and study of these scenarios is also part of this thesis.

Concept Learner for CorMeL Transaction Trees in Amadeus Data Processing Systems

Supervisor(s): Huang Xiao
Status: finished
Topic: Monitoring (VMI etc.)
Author: Heeren Sharma
Submission: 2014-02-26
Type of Thesis: Masterthesis
Proof of Concept No

Astract:

The architecture of Amadeus Data Processing Systems is distributed and highly coupled. To process an incoming request, various systems communicate with each other and process this request. CorMeL is a system which correlates all the logs gathered between those system components processing that incoming request. A request might occasionally not get processed due to changed configuration, hardware or software failures, and so on, and it is called as an Incident. In this thesis, approaches for anomaly detection in distributed information systems are analyzed and developed. This thesis also seeks to develop information retrieval methods from anomalous labeled data to locate possible failure causes for an incident in these systems. The results show that even though effectiveness of methods to locate failure causes is quite substantial, however, for certain scenarios, it demands deeper research. Discussion and study of these scenarios is also part of this thesis.