Job Management Node for Parallel Computing System
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In large-scale computer systems, fragmentation of free nodes can lead to inefficient operation and reduced throughput due to difficulties in quickly finding well-arranged groups of free nodes for job allocation, especially when erroneous job execution time estimates cause nodes to become fragmented, making it hard to secure sub-meshes with sufficient nodes for efficient communication.
Innovation Solution
A parallel computing system with a job management node that selects a job, searches for adjoining free nodes, determines a congruent area, and migrates jobs to create a well-arranged area for new job execution, optimizing node allocation and reducing the search range to improve processing speed and efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If job migration is performed throughout the entire system to find a well-arranged free node group, then a suitable migration destination can be found, but the search range becomes extensive and computation time increases enormously
Solution Approach 1:
The system divides the search space into multiple regions and performs job migration searches in parallel across these regions. The job management node segments the extensive search range into manageable portions, allowing simultaneous exploration of multiple potential migration destinations without requiring sequential examination of the entire system.
Solution Approach 2:
The system performs preliminary identification of free nodes and their spatial arrangements before initiating the full migration search process. By pre-processing and organizing information about available free nodes and their locations, the system reduces the computational burden during the actual migration decision-making phase.
2Adaptability or versatility
If free nodes are scattered in a fragmentation state, then node allocation flexibility is maintained, but it becomes difficult to secure a sub-mesh with sufficient nodes for efficient communication
Solution Approach 1:
The system dynamically adjusts job migration decisions based on real-time system state and future job requirements. Rather than maintaining static node allocations, the system continuously monitors free node distribution and performs targeted migrations to consolidate scattered free nodes into contiguous sub-meshes when needed, balancing flexibility with throughput requirements.
Solution Approach 2:
The system changes the spatial parameter of free node distribution by executing job migrations that transform scattered free node patterns into concentrated sub-mesh configurations. This parameter transformation enables the system to switch between maintaining allocation flexibility and achieving high throughput based on current needs.
Data Source
AI summary
A parallel-computing system includes: computation nodes connected in a mesh or torus connection and configured to perform parallel-computation; and a job management node including a computer, being coupled to the computation nodes, and being configured to: manage a plurality of jobs, select a first job from the jobs being executed by any of the computation nodes, search for a second area adjoining a first area occupied by a first computation node of the computation nodes executing the first job, the second area being occupied by a free computation node not executing any job, determine a fourth area of a same size as the first area at a location contacting an end in a predetermined direction in a third area comprising the first area and the second area, determine a fifth area of a predetermined shape in a range not overlapping the fourth area in the third area.