Toward clustering of network computing infrastructure objects based on analysis of statistical anomalies in network traffic

The problem of detecting statistical anomalies (that is, outliers in relation to the typical values of upload and download traffic) of the load on the nodes of the network computing infrastructure is considered. The regular scaling in computing resources and storage as well as redirection of data flows is needed due to the increase of load in real systems. The procedure for detecting statistical anomalies in network traffic is proposed using the approximation of observations by the generalized gamma distribution for further clustering of network computing infrastructure objects in order to evaluate resource need. All computational statistical procedures described in the paper are implemented using the R programming language and they are applied for network traffic, simulated using a specialized architectural and software stand. The proposed approaches can also be used for a wider class of telecommunication problems.