Server Selection for Carbon Emission Control

Abstract

Cloud owners are allowing their users to specify the level of resources being used in the di erent geographical locations that make up the cloud. The carbon emissions caused by powering these resources can vary greatly between di fferent geographical regions. The tra c for a given service can come from anywhere on the planet and the further the request has to travel the greater the negative e ect on quality of service (QoS). It is desirable to route tra ffic to the resources which cause the lowest carbon emissions but this can a ffect the QoS. A framework that characterizes this trade-o ff between carbon emissions and QoS is established in this paper. An algorithm that attempts to minimize the total cost of the trade-off described is presented. A traffi c generator is used to generate load for a server to establish functions which detail the carbon emissions and QoS of a service. We use these functions to simulate the performance of the algorithm in minimizing the total cost. Our results imply that carbon emissions can be reduced with little eff ect on the QoS under static tra ffic conditions and favourable energy supply conditions.