Cloud adoption is expanding rapidly as we witness a trend towards virtualization of resources such as hardware platforms, network resources, operating systems, etc. The main reasons are reduced capital expenditure, scalability, flexibility, reliability, and ease of upgrading. The cloud also provides an opportunity for its users to obtain on-demand computing services on a pay-as-you-go basis. However, the rapid expansion in the scale and complexity of cloud computing resources, cloud services, and application dependencies have made monitoring, controlling, and provisioning of resources more challenging and complex. Several solutions have been reported in the literature, but the problem is still of interest due to advances in methods, rapid increase in workload together with diverse and emerging application domains. As there are no first-principles dynamic models for cloud computing systems, the grey-box identification technique offers a possible alternative. Tools from control-theoretic techniques provide a rigorous mathematical foundation and systematic mechanisms to control and provision cloud computing resources.

In the first seminar, results presented on the identification of the web-server system and the design of a decentralized controller required continuous computing and updating of the control law.

In this talk, we present an event-triggered control formulation for determining the near-optimal number of computing resources required to serve the workload while guaranteeing performance as decided by the service level agreements. In an event-triggered control framework, the control law is updated only when an event of interest, such as deviation of the system from certain predefined performance metrics or violation of a predefined condition, has occurred. We derive conditions that ensure (i) stability of the closed-loop system under an event-triggered implementation and (ii) that the inter-event times are non-trivial. Finally, the validation and efficacy of the proposed approach on a real-time private cloud setup will be presented.