Stochastic Power Management for Data Center Workloads
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Data centers face significant power consumption challenges even during light workloads, as they maintain all systems running to minimize job completion time, leading to inefficiencies in power management.
Innovation Solution
Implementing stochastic power management (SPM) within data centers, where SPM control modules assess workloads and stochastically manage virtual machines (VMs) by migrating them between computer systems or allowing them to run to completion, thereby reducing the number of active systems and entering others into power-saving modes based on workload thresholds and network conditions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If jobs are distributed among all computer systems to minimize job completion time, then productivity is improved, but power consumption increases
Solution Approach 1:
The patent implements dynamic workload allocation where computer systems transition between active and power-saving states based on real-time workload conditions. The stochastic power management algorithm dynamically determines which systems remain active and which enter low-power modes, adjusting the distribution of jobs dynamically rather than statically assigning all systems to remain active.
Solution Approach 2:
The system changes the operational state parameter of computer systems from a binary always-on state to a variable state that can be either active or in power-saving mode. This parameter change is controlled by the stochastic algorithm that monitors workload intensity and adjusts system states accordingly, allowing the same hardware to operate in different power consumption regimes.
2Use of energy by moving object
If computer systems enter power saving mode to reduce power consumption, then energy efficiency is improved, but job completion time may increase
Solution Approach 1:
The stochastic power management algorithm performs preliminary assessment of workload intensity and system states before making power management decisions. By evaluating current conditions in advance and predicting future workload patterns, the system can proactively transition systems to power-saving modes when safe to do so, or pre-activate systems before workload increases, thereby avoiding delays in job completion.
Solution Approach 2:
The system implements feedback mechanisms where the stochastic algorithm continuously monitors job completion times, workload intensity, and system power states. This feedback loop allows the system to learn from past decisions and adjust power management strategies in real-time, ensuring that power-saving actions do not adversely impact job completion times while maximizing energy efficiency.
3Use of energy by moving object
If workload is concentrated on fewer systems to save power, then power consumption is reduced, but system reliability decreases
Solution Approach 1:
The patent applies local quality by allowing different computer systems to operate in different states simultaneously - some systems remain active while others enter power-saving modes. The stochastic algorithm determines the optimal local state for each individual system based on overall workload distribution, ensuring that sufficient systems remain active to maintain reliability while maximizing power savings where applicable.
Data Source
AI summary
Embodiments of the present disclosure describe methods, computer-readable media and system configurations for stochastic power management of one or more computer systems. A method may include ascertaining a workload of a plurality of computer systems (e.g., a data center). Additionally or alternatively, a method may include initiating, by a control module operated by a processor of a first of the plurality of computer systems, a stochastic power management process to manage power consumption of the first of the plurality of computer systems. The stochastic power management process may be conditionally initiated based at least in part on the ascertained workload of the plurality of computer systems. The stochastic power management process may include a plurality of virtual machine management actions having corresponding probabilities being taken, one or more of which may result in power savings. Other embodiments may be described and/or claimed.