Datacenter Load Shedding by Workload-Aware Response Levels
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional datacenter power management techniques for load shedding are suboptimal, as they lack insight into workloads and customer impacts, leading to inefficient power reduction and potential broader disruptions, and often result in excessive or insufficient power reductions.
Innovation Solution
A method that identifies and prioritizes workloads on hosts within a datacenter, selects appropriate response levels for power reduction actions based on estimated impact and power reduction, and applies these actions dynamically to minimize disruption while maintaining power balance, using a computer system to orchestrate power consumption reductions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of energy
If manual load shedding is performed by shutting down host racks or individual devices one by one, then power consumption is reduced, but the operator has no insight into workloads or customers impacted and the approach is suboptimal
Solution Approach 1:
The system implements automated feedback loops that continuously monitor power consumption levels, workload status, and customer impact metrics. This feedback enables the load shedding system to make informed decisions about which workloads to migrate or terminate, ensuring that power reduction actions are based on real-time data about system state and customer priorities rather than manual estimation.
Solution Approach 2:
The system enables workloads to be automatically migrated or terminated based on pre-configured policies and real-time conditions, without requiring continuous manual operator intervention. The automated system serves itself by monitoring its own power consumption, evaluating available actions, and executing appropriate load shedding decisions based on workload priorities and customer impact assessments.
2Loss of energy
If emergency shut off switch is pressed to turn off power to the whole or large portion of the datacenter, then power consumption is rapidly reduced, but the impact on customers and workloads is excessive
Solution Approach 1:
The system segments the datacenter into individual host racks and further into specific workloads, enabling granular control over load shedding actions. Instead of applying power reduction uniformly across the entire datacenter, the system can selectively target specific hosts or workloads based on their power consumption, priority levels, and customer impact, thereby achieving power reduction with minimized disruption.
Solution Approach 2:
The system applies different load shedding strategies to different parts of the datacenter based on local conditions such as workload priority, customer SLAs, and power consumption characteristics. High-priority workloads may be protected from load shedding while lower-priority workloads are targeted first, ensuring that power reduction actions are tailored to local requirements rather than applied uniformly.
3Loss of energy
If conventional load shedding methods are used, then power consumption is reduced, but the power reduction may be insufficient or excessive without proper estimation
Solution Approach 1:
The system performs preliminary assessments of available power reduction actions and their expected impact before executing load shedding. By evaluating the power consumption characteristics of different workloads and predicting the outcome of potential actions, the system can determine the appropriate level of load shedding needed to achieve the desired power reduction target, avoiding both insufficient and excessive power reduction.
4Measurement precision
If automated systems are implemented to improve load shedding precision, then power reduction estimation improves, but device complexity increases
Solution Approach 1:
The system employs a multi-functional platform that combines power monitoring, workload management, migration capabilities, and load shedding decision-making in a single integrated system. This universal approach avoids the need for multiple separate systems and reduces overall complexity by consolidating functions that work together to achieve precise power reduction estimation and execution.
Data Source
AI summary
Disclosed techniques relate to orchestrating power consumption reductions across a number of hosts. A number of response levels may be utilized, each having an association to a corresponding set of reduction actions. The impact to customers, hosts, and/or workloads can be computed at run time based on current and/or predicted conditions and workloads, and a particular response level can be selected based on the computed impact. These techniques enable a sufficient, but least impactful response to be employed.