ONLINE DETECTION OF ICT SERVER FAILURE ON DEMAND
ICT servers continually record information regarding load, number of requests, ping time, etc. Usually, for detecting abnormal states, the collected data are compared to a fixed threshold: this can lead to false and missed alarms which can result in the large unavailability of the ICT server.
To overcome this problem, Aramis has resorted to statistical and machine learning techniques for the definition of a dynamic and online updating threshold for fault detection: this threshold has been computed considering the last collected data exploiting a specifically developed unsupervised framework.
The developed adaptive threshold allows dynamically identifying the server faults taking into account the effects of the continuously varying load request.