Edge Cluster Autoscaling Using Power-Aware Node Activation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing edge computing systems fail to consider power metrics during autoscaling, leading to inefficient resource utilization and increased energy consumption, particularly in resource-constrained edge clusters powered by renewable energy sources.
Innovation Solution
Implement adaptive cloud autoscaling that incorporates power considerations, such as battery levels and renewable energy availability, to dynamically adjust node activation, deactivation, and resource allocation in edge clusters, ensuring optimal power settings without compromising Service Level Agreements (SLAs).
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Use of energy by moving object
If traditional autoscaling is used without power metrics, then resource allocation responds to compute load, but energy consumption increases and renewable energy opportunities are missed
Solution Approach 1:
The system implements feedback loops that continuously monitor power metrics (battery levels, renewable energy availability) and compute load, then use this feedback to dynamically adjust autoscaling decisions. This closed-loop control enables the system to respond to both computational needs and power availability, resolving the contradiction between energy efficiency and resource utilization.
Solution Approach 2:
The autoscaling system transitions from static threshold-based scaling to dynamic power-aware scaling. Power metrics are integrated as variable parameters that continuously influence scaling decisions, allowing the system to adapt resource allocation in real-time based on both compute demand and power availability conditions.
2Reliability
If edge nodes are continuously active to meet SLAs, then service reliability is maintained, but energy consumption increases
Solution Approach 1:
The system performs preliminary assessments of power availability and workload characteristics before making scaling decisions. By predicting future power states and workload patterns, the system can proactively scale resources to meet upcoming SLA requirements during periods of adequate power availability, avoiding the need for continuous node activation.
Solution Approach 2:
The system dynamically changes operational parameters (node activation states, resource allocation levels) based on real-time power metric thresholds. When power availability exceeds thresholds, additional nodes are activated to meet SLAs; when power is constrained, non-critical nodes are deactivated, maintaining reliability only where power is available.
3Loss of energy
If power metrics are integrated into autoscaling, then energy efficiency improves, but system complexity increases
Solution Approach 1:
The autoscaling system is enhanced to serve multiple functions simultaneously: traditional compute-load-based scaling and power-aware scaling. The same control mechanism processes both types of inputs and generates unified scaling decisions, avoiding the need for separate complex subsystems and reducing overall system complexity despite the added power management capability.
4Object-generated harmful factors
If edge computing is deployed at remote locations with renewable power, then carbon emissions are reduced, but power availability becomes unreliable
Solution Approach 1:
The system dynamically adapts compute operations to the variable power supply conditions of remote renewable energy sources. By continuously monitoring power availability and adjusting resource allocation accordingly, the system maintains operational reliability despite the inherent intermittency of renewable power sources at edge locations.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
Various aspects of methods, systems, and use cases include using an edge device to control power states of nodes in an edge cluster. A method may include scheduling unscheduled application pods on a set of nodes of an edge cluster. The method may include determining a change to at least one power state of the set of power states based on the schedule. The method may include causing the edge cluster to operate according to the change.