VM Power Scheduling Framework for Data Center Energy Optimization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Data centers face high operating costs and carbon footprints due to energy inefficiencies, particularly during varying workloads, as servers are often provisioned for peak loads rather than average loads, leading to low utilization and significant energy waste.
Innovation Solution
A power scheduling framework for virtual machines that dynamically adjusts their power modes based on workload, optimizing the number of active servers and their power states to minimize energy consumption while maintaining service quality, by considering transition energy costs and using dynamic voltage and frequency scaling.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If servers are provisioned for peak loads, then service reliability is improved, but energy consumption increases
Solution Approach 1:
The patent implements dynamic power mode adjustment for virtual machines based on real-time workload conditions. The system continuously monitors service instance arrival rates and dynamically transitions VMs between active and lower power modes, making the infrastructure adaptive rather than static. This resolves the contradiction by provisioning capacity dynamically - having full capacity available when needed (maintaining reliability) but operating at reduced capacity during low-demand periods (reducing energy consumption).
Solution Approach 2:
The system changes the power mode parameter of virtual machines based on workload conditions. By adjusting the power state parameter (active, lower power mode, or suspended) according to service instance arrival rates, the system optimizes the balance between service reliability and energy consumption. This parameter adjustment allows the same physical infrastructure to serve different operational requirements without permanent over-provisioning.
2Use of energy by moving object
If virtual machines are transitioned between power modes, then energy savings are achieved, but transition energy cost is incurred
Solution Approach 1:
The system performs preliminary actions by suspending virtual machines to a lower power mode in advance during low-demand periods, rather than allowing them to remain active and then shutting them down abruptly when needed. This preliminary transition to a lower power state reduces the energy cost of subsequent power state changes and minimizes wasted transition energy, as the VM is already in an optimized state rather than requiring a full power cycle.
Solution Approach 2:
The system implements periodic monitoring and adjustment of virtual machine power modes based on service instance arrival patterns. By periodically evaluating workload conditions and adjusting power states accordingly, the system optimizes the timing and frequency of transitions, ensuring that transition energy costs are incurred only when necessary and that energy savings are maximized through rhythmic optimization cycles.
3Reliability
If servers run continuously, then service availability is maintained, but utilization decreases
Solution Approach 1:
The patent segments the virtual machine population into different power state groups (active, lower power mode, suspended) based on workload demands. This segmentation allows the system to maintain service availability through the active group while optimizing utilization by placing excess capacity in lower power states. The segmentation enables independent management of different VMs based on actual service needs rather than treating all servers uniformly.
Solution Approach 2:
The system implements a universal power management framework that can apply the same dynamic power mode adjustment logic across all virtual machines regardless of their specific workloads or service requirements. This multi-functional approach allows a single infrastructure to serve multiple operational modes (high availability, energy savings, balanced operation) by universally applying the workload-based power optimization algorithm across the entire VM portfolio.
Data Source
AI summary
Provision of power scheduling framework for a plurality of virtual machines to facilitate energy savings in handing service instances includes configuring a first set of virtual machines in an active power mode and a second set of virtual machines in one or more other power modes, and managing, across a time period and based on service instance arrival, power mode configuration for the first set of virtual machines and the second set of virtual machines, the managing including initiating one or more transitions of at least one virtual machine between power modes of the plurality of power modes, wherein a transition energy cost is associated with transitioning from a lower power mode of the plurality of power modes to a higher power mode of the plurality of power modes, and wherein the managing accounts for transition energy cost in determining the one or more transitions.