Scalable monitoring and optimization techniques for mega-scale data centers

This paper focuses on the design of service provisioning schemes suitable for mega data center (DC) infrastructures. A major issue linked with the operation of these infrastructures is scalability caused by the increased number of resources available in mega-size highly-dense DCs and the associated requirements for control and management information. To address this scalability issues, we propose for the first time to monitor and optimize the operation of mega DCs adopting graph factorization combined with compressive sensing theories. This approach takes advantage of the spatial and temporal correlation of compute, and network resource requests, to monitor and optimize metrics, such as delay and energy with reduced control and management information. Our modelling results indicate drastically reduced volume of traffic transferred from the data to control plane and number of optimization process variables.