Resource Scheduling via Coordinate Space Caching
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional resource scheduling methods for computer clusters and cloud computing platforms are inefficient due to high computational complexity and inability to effectively handle diverse resource requirements and locality constraints, leading to time-consuming task allocation processes.
Innovation Solution
A resource scheduling method that maps nodes to a coordinate space with leaf buckets, allowing for caching of search results to accelerate the selection of suitable nodes based on requested resource values and locality constraints, reducing computational complexity through positive and negative caching strategies.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional resource scheduling methods are used to select nodes for tasks, then all candidate nodes are evaluated to ensure optimal assignment, but the computational complexity becomes prohibitive and scheduling time increases significantly
Solution Approach 1:
The patent applies preliminary action by pre-computing and storing node compatibility information in a coordinate space before actual task scheduling occurs. Nodes are mapped to coordinate buckets based on their resource characteristics in advance, so when a task arrives, the scheduler only needs to query pre-organized data rather than evaluating all nodes from scratch. This resolves the contradiction by preparing the evaluation framework beforehand, reducing real-time computational complexity while maintaining accurate node selection.
Solution Approach 2:
The patent segments the node selection process by dividing the coordinate space into discrete buckets based on resource dimensions. Instead of evaluating nodes in a continuous space, the system partitions nodes into discrete categories (buckets) based on their resource profiles. This segmentation allows the scheduler to quickly identify relevant buckets for a given task without examining every node individually, thus reducing computational complexity while preserving selection accuracy through structured organization.
2Measurement precision
If all candidate nodes are evaluated for each task to ensure proper resource matching, then node selection accuracy is maintained, but the scheduling throughput decreases due to time-consuming evaluation processes
Solution Approach 1:
The system performs preliminary organization of nodes into coordinate buckets based on their resource characteristics before task arrival. This pre-computation allows the scheduler to quickly retrieve candidate nodes from pre-filtered buckets rather than evaluating all nodes for each task. The preliminary structuring of data maintains accurate resource matching while dramatically improving scheduling throughput by reducing the evaluation scope for each incoming task.
Solution Approach 2:
The patent creates a virtual coordinate space representation (a copy) of the physical node pool. Instead of directly querying and evaluating actual nodes for each task, the scheduler operates on this virtual coordinate model where nodes are already organized by resource characteristics. This copying approach maintains accurate resource matching decisions while improving throughput by enabling faster queries on the virtual structure compared to direct node evaluation.
3Reliability
If comprehensive node evaluation is performed considering all resource dimensions and constraints, then task assignment quality is improved, but the time required for each scheduling decision increases
Solution Approach 1:
The patent segments the comprehensive node evaluation process by organizing nodes into discrete coordinate buckets based on resource dimensions. Each bucket represents a segment of the resource space, allowing the scheduler to evaluate only relevant segments rather than the entire node pool. This segmentation maintains task assignment quality by ensuring all relevant constraints are considered within the appropriate buckets while reducing scheduling decision time through localized evaluation.
Solution Approach 2:
The system performs preliminary organization of nodes into coordinate buckets that encode resource dimension information. This pre-computed structure allows the scheduler to quickly identify and evaluate only the relevant node segments for each task based on its specific requirements. The preliminary structuring ensures comprehensive constraint consideration is maintained while dramatically reducing the time required for each scheduling decision by limiting evaluation to pre-identified candidate segments.
Data Source
AI summary
Resource scheduling methods and a resource manager define leaf buckets, each of which have coordinates in corresponding dimensions in a coordinate space. Nodes having resource values in the corresponding dimensions are mapped on the leaf buckets. A task request identifies candidate nodes and requested resources that relate to the coordinate space dimensions. The leaf buckets are scanned to find a fittable node among the list of candidates. A result of the scan is stored in a positive or a negative cache entry. When a next task request of a same task group is received, scanning of the leaf buckets is performed using information from the positive or negative cache entry. For a positive cache entry, the scanning is initiated from a point where the previous scanning has ended. For a negative cache entry, the scanning ignores nodes that were found not to contain the requested resources in the previous scanning.