Adaptive Scale-In Threshold for Cloud Cluster Resource Reuse
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current automatic scaling features in cloud computing environments do not effectively minimize the use of costly on-demand host computers, as they do not consider the associated costs during scaling operations.
Innovation Solution
Implementing an aggressive scale-in resource utilization threshold to identify target clusters for resource reduction and reuse reserved resources instead of adding on-demand resources, thereby optimizing the use of reserved host computers.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If automatic scaling adds on-demand host computers when reserved instances are exhausted, then cluster resource availability is improved, but cloud computing costs increase
Solution Approach 1:
The system dynamically changes the scale-in threshold parameter from a fixed value to an adaptive value that increases when reserved instance availability drops below a threshold. This parameter change enables the system to prioritize cost-effective reserved instances over on-demand instances by being more aggressive about scaling in clusters with reserved instances, thereby reducing cloud computing costs while maintaining resource availability through strategic use of on-demand instances only when necessary
Solution Approach 2:
The system implements a feedback mechanism that monitors the availability of reserved resource instances and uses this information to adjust scaling behavior. When reserved instances become scarce, the system receives feedback about this condition and responds by increasing scale-in thresholds to free up reserved instances for future use, creating a closed-loop control system that balances resource availability with cost optimization
2Productivity
If standard scale-in thresholds are used to remove host computers from clusters, then resource utilization is optimized, but reserved resource instances are not recovered in time for scaling operations
Solution Approach 1:
The system performs preliminary action by proactively increasing the scale-in threshold before reserved instances are completely exhausted. This anticipatory adjustment allows the system to begin recovering reserved instances earlier in the utilization cycle, creating a buffer of available reserved instances before they are needed for scaling operations, thereby reducing the time loss associated with instance recovery
3Reliability
If aggressive scale-in operations are performed to recover reserved instances, then availability of reserved resources is improved, but cluster stability may be compromised
Solution Approach 1:
The system applies dynamics by making the scale-in threshold adaptive rather than static. The threshold dynamically adjusts based on the availability of reserved instances, being more aggressive (higher threshold) when reserved instances are scarce and less aggressive (lower threshold) when they are abundant. This dynamic adjustment allows the system to perform aggressive scale-in operations only when necessary, thereby recovering reserved resources while maintaining cluster stability through context-appropriate scaling behavior
Data Source
AI summary
System and computer-implemented method for autoscaling clusters of host computers in a cloud-based computing environment uses an aggressive scale-in resource utilization threshold that is greater than a corresponding standard scale-in resource utilization threshold to search for any target clusters of host computers in response to a scale-out recommendation for a cluster of host computers to select a candidate cluster of host computers when the number of available reserved resource instance for the cloud-based computing environment is below a predefined value. A scale-in operation is executed on the candidate cluster of host computers to remove an existing resource instance from the candidate cluster of host computers. A scale-out operation is executed on the cluster of host computers using an available resource instance for the cloud-based computing environment.