Monitoring and optimization of ATLAS Tier 2 center GoeGrid

The demand on computational and storage resources is growing along with the amount of information that needs to be processed and preserved. In order to ease the provisioning of the digital services to the growing number of consumers, more and more distributed computing systems and platforms are actively developed and employed. The building block of the distributed computing infrastructure are single computing centers, similar to the Worldwide LHC Computing Grid, Tier 2 centre GoeGrid. The main motivation of this thesis was the optimization of GoeGrid performance by efficient monitoring. The goal has been achieved by means of the GoeGrid monitoring information analysis. The data analysis approach was based on the adaptive-network-based fuzzy inference system (ANFIS) and machine learning algorithm such as Linear Support Vector Machine (SVM). The main object of the research was the digital service, since availability, reliability and serviceability of the computing platform can be measured according to the constant and stable provisioning of the services. Due to the widely used concept of the service oriented architecture (SOA) for large computing facilities, in advance knowing of the service state as well as the quick and accurate detection of its disability allows to perform the proactive management of the computing facility. The proactive management is considered as a core component of the computing facility management automation concept, such as Autonomic Computing. Thus in time as well as in advance and accurate identification of the provided service status can be considered as a contribution to the computing facility management automation, which is directly related to the provisioning of the stable and reliable computing resources. Based on the case studies, performed using the GoeGrid monitoring data, consideration of the approaches as generalized methods for the accurate and fast identification and prediction of the service status is reasonable. Simplicity and low consumption of the computing resources allow to consider the methods in the scope of the Autonomic Computing component.