Dynamic Health Score Workload Scheduling
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current workload scheduling in compute environments is inefficient due to the lack of consideration for network state, relying on instantaneous statistics and static weights, which leads to suboptimal scheduling decisions and frequent workload migrations.
Innovation Solution
A method that calculates a health score for network elements using machine learning techniques, incorporating temporal and spatial statistics to assign dynamic weights, enabling more accurate scheduling decisions by considering the health of ports, switches, and the entire network fabric.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If current workload scheduling algorithms use static weights and instantaneous statistics for health computation, then the scheduling process is simple and fast, but the health evaluation accuracy is low leading to suboptimal scheduling decisions
Solution Approach 1:
The patent applies dynamics by transitioning from static weights to dynamic weights that adapt over time. The system uses temporal statistics to capture how network health metrics evolve, allowing the scheduling algorithm to adjust weights based on changing conditions rather than relying on fixed predetermined values.
Solution Approach 2:
The patent introduces spatial characteristics as an additional dimension to the health evaluation. By incorporating spatial statistics that consider the topological relationships and physical layout of network elements, the system moves beyond simple instantaneous measurements to a multi-dimensional assessment that includes both temporal evolution and spatial context.
2Reliability
If workload scheduling considers only instantaneous network statistics, then the scheduling decision is made quickly, but frequent workload migrations occur due to inaccurate health assessment
Solution Approach 1:
The patent applies preliminary action by using temporal statistics to predict future network health states before making scheduling decisions. Instead of reacting to current instantaneous conditions alone, the system analyzes historical trends and patterns to anticipate potential issues, allowing workloads to be placed on network elements that are likely to remain healthy.
Solution Approach 2:
The system implements feedback mechanisms by continuously monitoring network health metrics and using this information to refine scheduling decisions. The temporal and spatial statistics provide ongoing feedback about network conditions, allowing the scheduler to adapt and improve its decisions over time rather than making isolated one-time choices.
3Measurement precision
If the scheduling algorithm uses comprehensive temporal and spatial statistics with machine learning, then scheduling accuracy improves, but the computational overhead increases
Solution Approach 1:
The patent applies partial action by selectively applying machine learning techniques to the most critical or uncertain aspects of health evaluation rather than processing all possible statistics with equal computational intensity. The system focuses computational resources on temporal and spatial features that have the greatest impact on scheduling decisions.
Solution Approach 2:
The system employs self-service through machine learning models that automatically learn and adapt from the collected temporal and spatial statistics without requiring manual intervention or extensive computational optimization. The algorithms self-adjust to capture patterns in the data, reducing the need for complex preprocessing or manual feature engineering.
Data Source
AI summary
Disclosed is a method that includes collecting first temporal statistics for a port element in a computing environment, collecting second temporal statistics for a switch element in the computing environment, collecting third temporal statistics for the computing environment generally, computing a spatial correlation between network features and network elements comprising the port element and the switch element and computing, via a machine learning technique, a port dynamic weight for the port element and a switch dynamic weight for the switch element. The method can also include scheduling workload to consume compute resources within the compute environment based at least in part on the port dynamic weight for the port element and the switch dynamic weight for the switch element.