Cluster Node Selection Using Off-Band Health Metrics
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing computing cluster management systems lack awareness of off-band metrics, leading to suboptimal node selection that compromises operational efficiency and reliability during dynamic adjustments to evolving workloads.
Innovation Solution
Implementing a system that utilizes both on-band and off-band metrics to determine utilization, reliability, and vulnerability scores for nodes, enabling a trained controller/scheduler to make informed decisions on node addition and removal based on comprehensive system health and security assessments.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If only on-band metrics are used for node selection, then the system complexity is reduced and ease of operation is improved, but the reliability and measurement precision of node assessment deteriorates
Solution Approach 1:
The patent segments metrics collection into two distinct bands: on-band metrics (collected through standard OS-level interfaces) and off-band metrics (collected through dedicated hardware management controllers). This segmentation allows the system to gather comprehensive node health information without overwhelming the control plane, as each band operates through its own established protocols and interfaces.
Solution Approach 2:
The patent introduces an intermediary component that aggregates and processes metrics from both on-band and off-band sources. This intermediary synthesizes the raw data into a unified node health score, simplifying the decision-making process while maintaining comprehensive assessment capabilities. The intermediary acts as a buffer between the complex metrics collection infrastructure and the node selection logic.
2Measurement precision
If comprehensive metrics collection is implemented, then the measurement precision and reliability of node assessment is improved, but the loss of time and productivity decreases due to increased processing overhead
Solution Approach 1:
The patent implements preliminary action by pre-collecting and caching off-band metrics from hardware management controllers before node selection is needed. These metrics are gathered in advance and stored in a readily accessible format, so when a node selection decision is required, the system can quickly retrieve pre-processes data without initiating time-consuming collection operations at the moment of decision-making.
Solution Approach 2:
The patent transforms multiple diverse metrics from different sources into a standardized parameter format with uniform data types and scales. By normalizing metrics into consistent parameters, the system enables efficient comparison and aggregation operations, reducing the computational time required to assess node health while maintaining comprehensive evaluation precision.
Data Source
AI summary
A system can determine respective on-band metrics for respective nodes of a group of nodes that are part of a computing cluster, wherein a first part of the group of nodes are active to the computing cluster, and wherein a second part of the group of nodes are inactive to the computing cluster. The system can determine respective off-band metrics for the respective nodes. The system can determine respective utilization scores of the respective nodes based on the respective on-band metrics. The system can determine respective reliability scores of the respective nodes based on the respective off-band metrics. The system can determine respective vulnerability scores of the respective nodes based on the respective on-band metrics. The system can modify a number of nodes of the first part of the group of nodes based on the respective utilization scores, the respective reliability scores, and the respective vulnerability scores.


