AI Workload Scheduler for Cloud Node Affinity Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current workload scheduling in cloud computing systems relies on static and predefined rules, which fail to adapt dynamically to the changing characteristics of workloads, leading to inefficient resource allocation and increased costs due to underutilization of resources.
Innovation Solution
An AI-driven workload scheduler that uses machine learning to evaluate both static and dynamic characteristics of workloads, converting them into feature vectors and performance indices to predict optimal node affinity, thereby dynamically scheduling workloads across cloud computing resources.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If static and predefined rules are used for workload scheduling, then implementation simplicity is maintained, but resource allocation efficiency deteriorates due to inability to adapt to changing workload characteristics
Solution Approach 1:
The patent transforms the static scheduling system into a dynamic one by implementing machine learning models that continuously learn from historical workload performance data and adapt to changing workload characteristics. The system dynamically updates cloud affinity factors based on real-time performance metrics, enabling adaptive resource allocation while maintaining systematic control.
Solution Approach 2:
The patent implements feedback mechanisms where workload performance data is collected, analyzed, and used to update scheduling decisions. The machine learning models continuously receive feedback from actual workload execution results and adjust cloud affinity factors accordingly, creating a closed-loop system that improves resource allocation efficiency over time.
2Productivity
If machine learning models are used to evaluate workload characteristics, then resource utilization improves through intelligent scheduling, but system complexity increases due to additional processing requirements
Solution Approach 1:
The patent performs preliminary actions by pre-training machine learning models with historical workload data before actual scheduling operations. The system pre-calculates cloud affinity factors based on static workload characteristics and historical performance patterns, so that during runtime, scheduling decisions can be made quickly without extensive real-time computation.
Solution Approach 2:
The patent introduces an intermediary layer consisting of machine learning models that sit between the workload submission and the scheduling decision. These models act as mediators that translate raw workload characteristics into meaningful cloud affinity factors, simplifying the overall scheduling process while improving resource utilization through intelligent analysis.
3Measurement precision
If dynamic characteristics are analyzed in real-time, then scheduling accuracy improves, but processing time increases due to additional evaluation requirements
Solution Approach 1:
The patent performs preliminary analysis of workload static characteristics before runtime scheduling. Machine learning models are pre-trained on historical performance data, and cloud affinity factors are pre-calculated based on static workload attributes. This preliminary action enables fast real-time scheduling decisions without extensive processing during execution.
Solution Approach 2:
The patent implements a hybrid approach where static workload characteristics are analyzed in full detail during preprocessing, while only critical dynamic characteristics are monitored in real-time. This partial real-time analysis maintains scheduling accuracy for the most important factors while minimizing processing time overhead during runtime.
Data Source
AI summary
Mechanisms are provided for scheduling a workload in a cloud computing system. A cloud affinity factor (CAF) computer model is trained, via a machine learning process based on a training dataset comprising static characteristics of a workload binary for a workload, and dynamic characteristics corresponding to historical performance data for the workload, such that the trained CAF computer model predicts a performance classification for a given workload binary. The trained CAF computer model processes a new workload to generate a performance classification for the new workload. Cloud affinity factor(s) are generated based on the performance classification for the new workload. Node affinity and dispatch rule(s) are applied to the cloud affinity factor(s) to select one or more nodes of the cloud computing system to which to dispatch the workload. The workload is then scheduled on the selected one or more nodes.