Monitoring heterogeneous systems and implementing error prediction methods

OData support
Dr. Sonkoly Balázs
Department of Telecommunications and Media Informatics

The complex, multiple-component heterogeneous systems more errors also arise through a number of errors and spill over, causing an impact on other devices, services, applications. It would be nice to know in advance preach that the number of errors in the future when they may occur.

I presented the monitored system and components, as well as the necessary tools. I compared the Nagios, the NetIQ, the ManageEngine and the LogicMonitor each other, then chosen the Nagios software tools, and described in more it.

The selected monitoring device components in the network monitors, and information about errors that status can be retrieved from the system. To achieve this, first, I used the HtmlUnit, which is operated inefficiently, so I found a better solution is to LiveStatus. This is a mudol I switched to Nagios and LQL (Livestatus Query Language) query language to retrieving the necessary information about the monitoring device.

I presented a classification, clustering algorithms, which are extracted from the classified system errors. I chosem the k-means clustering algorithm and and described in more it. Then I presented the frequent sequential patterns search algorithms and selected the SPAM method, which I have explained more.

SPAM and the k-means algorithm used the implementation of classification errors. The former requires the data sequences, and the latter vector the fore. Clusters are implemented using the error prediction algorithms.

Dependencies in the system is stored in a directed graph so that each component was considered as a node, and the correlations (eg message exchange) between them the edges. Then, the dependency directed graph based fault prediction is implemented.

The resulting error statistics groups made the mistake and correct operation I tested the prediction algorithms.


Please sign in to download the files of this thesis.