A Cross-Domain Energy Optimization Approach for End-to-End Cloud Environments

Cloud services emerged to provide easy and affordable access to computing resources over the Internet and doing so eliminated the need for managing the complex in-house IT environment. Data centers consume a significant amount of power. According to a report, in 2020 the data center (DC) industry consumed around 196 to 400 terawatt-hours. In half a decade, the projected data center power consumption usage is going to be around 33% of total global electricity consumption which may be 20% of the entire energy consumed around the world. Cloud service spends a significant portion of its operational expense toward electricity usage. Therefore, efficient utilization of electricity in the DC sites is essential to reduce costs, energy consumption, carbon emissions and environmental impact for a sustainable cloud service which hinges heavily on efficient use of power consumed by the cooling system and the servers at different data centers in the cloud. The proposed system AYPOS, optimizes the Power Usage Effectiveness (PUE) of the overall cloud infrastructure by dynamically changing the virtual machine (VM) topology (by moving clusters of VMs within or across DC sites) to de-activate unnecessary IT components and thereby lowering cooling power utilization and tuning the DC Information Management (DCIM) in a cross-domain manner. The interrelations between the power consumed by the cooling system (aka the passive system) and that of IT components (aka active system) in each site must be properly modeled to predict and control data center power efficiency. One of the tools recently emerged for efficient operation of data centers is ‘virtualization’ that encompasses a wide array of technologies that creates slices of the physical computing, storage and networking that can be easily moved around and software-reconfigured. A virtual machine (VM) which is a partition of a physical computer that has all the functions of a server including its own operating system, improves the efficiency of data centers significantly; but the optimal placement/assignment of VMs to physical machines (servers) is a big challenge. The migration of VM workload from time to time among data sites has become an opportunity to provide cloud reconfiguration, and hence a great tool for power management. While there are many variables to consider, the focus of this work is PUE reduction across a DC to optimally move VMs by modeling the complex relationship between the power usage of IT equipment and cooling equipment. The task of moving VMs requires an enormous amount of data to be collected and stored from all IT equipment, cooling systems, and meteorological systems. The mathematical relationship between these components that contribute to overall data center power is complicated, time-varying, and nonlinear. A system and several methods are presented for an end-to-end cloud with multi-data-center (DC) infrastructure management wherein the service is offered to a plurality of users by a plurality of virtual machines (VM)s configured on each physical machine, each VM having a certain set of service level agreements (SLAs) and usage pattern. In this paper, a general architecture is presented for a cross-domain approach for the power utilization efficiency in a E2E Cloud environment; the details of the algorithms are pending for patent.