Dynamic Cloud Resource Allocation Against Oversubscription
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Cloud-based resource allocation is complex due to multiple logical server systems sharing hardware, making it difficult to control hardware resource utilization and predict potential oversubscription, which can lead to resource exhaustion.
Innovation Solution
A resource allocation system that uses machine learning to analyze historical utilization data and generate warnings for oversubscribed resource classes, identifying supplemental classes to prevent oversubscription by dynamically adjusting resource allocation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If multiple logical server systems are hosted on a single hardware system, then resource utilization efficiency is improved, but control over hardware resource utilization becomes more difficult
Solution Approach 1:
The patent segments hardware resources into multiple virtual server instances, allowing independent management and control of each virtual instance while sharing underlying hardware resources. This enables fine-grained control over resource allocation to different logical servers, resolving the contradiction between improved utilization efficiency and difficulty of control.
Solution Approach 2:
The patent implements feedback mechanisms that continuously monitor hardware resource utilization across multiple virtual servers and automatically adjust resource allocation. This closed-loop control system maintains ease of operation by automating resource management while maximizing utilization efficiency through dynamic adjustments based on actual usage patterns.
2Reliability
If different classes of hardware systems are used to guarantee performance levels, then service quality is improved, but resource allocation complexity increases
Solution Approach 1:
The patent creates a universal virtualization layer that can accommodate multiple performance guarantee levels within a single hardware system. This multi-functional platform provides different service quality levels (guaranteed, best-effort, reserved) without requiring separate hardware classifications, thereby reducing allocation complexity while maintaining service quality differentiation.
Solution Approach 2:
The patent uses parameter changes in resource allocation policies to differentiate performance guarantee levels. By adjusting allocation parameters (such as priority weights, reservation amounts, and throttling thresholds) rather than requiring different hardware classes, the system achieves varied service quality levels without increasing hardware system complexity.
3Productivity
If resources are oversubscribed to maximize utilization, then resource efficiency is improved, but risk of resource exhaustion increases
Solution Approach 1:
The patent applies preliminary action by establishing resource reservation mechanisms and predictive monitoring before resource exhaustion occurs. The system proactively identifies trends toward oversubscription and takes preventive measures (such as allocating supplemental resources or adjusting allocations) to maintain resource availability while preserving efficiency through controlled oversubscription.
Solution Approach 2:
The patent implements dynamic resource allocation that continuously adjusts oversubscription levels based on current utilization patterns and predicted demand. This dynamic approach allows the system to maximize resource efficiency during low-demand periods while automatically reducing oversubscription and preventing exhaustion during high-demand periods, thereby maintaining both productivity and reliability.
Data Source
AI summary
Methods and systems are described herein for a resource allocation system. The resource allocation system may obtain a corresponding quantity of resources (e.g., memory, processor, storage, etc.) needed to be allocated for each resource class (e.g., for a given performance class) for a particular time period (e.g., for one month). Furthermore, the resource allocation system may track allocation of each class of resources and may predict that some classes of resources will be oversubscribed. Based on the prediction, the resource allocation system may, using a machine learning model, identify supplemental classes for each resource class predicted to be oversubscribed and generate a warning when a resource of a supplemental class is predicted to be used.