Analysis of the macroscopic behavior of server systems in the internet environment

Elasticity is one of the key features of cloud-hosted services built on virtualization technology. To utilize the elasticity of cloud environments, administrators should accurately capture the operational status of server systems, which changes constantly according to service requests incoming irregularly. However, it is difficult to detect and avoid in advance that operating services are falling into an undesirable state. In this paper, we focus on the management of server systems that include cloud systems, and propose a new method for detecting the sign of undesirable scenarios before the system becomes overloaded as a result of various causes. In this method, a measure that utilizes the fluctuation of the macroscopic operational state observed in the server system is introduced. The proposed measure has the property of drastically increasing before the server system is in an undesirable state. Using the proposed measure, we realize a function to detect that the server system is falling into an overload scenario, and we demonstrate its effectiveness through experiments.