2016 Forum – Methodology and Approach to Estimating the Power Consumption of a Workload

09 March, 2016 | Presentation

Jay Vincent, Intel Corporation

To become more power efficient in a dynamic workload environment indicative of cloud computing and software-defined “X,” we need to better understand how processes, services, virtual machines, or containers consume power. This can become a complex problem in a dynamic resource pool where there are many compute nodes, cores, and virtual instances in a multi-tenant environment. One approach is to use platform-based power sensors and combine them with resource monitors to determine the essential power load of a compute task. It will be a futile effort to drive power efficiency and optimize cost in a compute environment without having visibility into how the workloads consume power at the compute node. Until we can better manage the demand for power, we will never be able to eliminate the waste associated with supplying the power. This paper provides a technical overview of how to understand the power profile of the workloads that make up the demand for power in a data center.