Job Management Device for Parallel Resource Search Optimization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In large parallel computer systems, the efficiency of job scheduling is hindered by the complexity of resource management due to the exponential growth of wiring and relay devices, leading to inefficiencies in processor assignment and communication traffic, which results in reduced system performance and throughput.
Innovation Solution
A method utilizing a graph representation of the computer system with vertices and edges to manage computing resources, introducing concepts like 'opposite shores' and 'n-th nearest neighbors' to optimize resource assignment and search processes, and calculating a complexity parameter to estimate search times accurately.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If a mesh or torus network is used to suppress the quantity of wirings and relay devices, then the device complexity is reduced, but the job scheduling process time increases due to complex resource management
Solution Approach 1:
The patent divides the search space for resource allocation into multiple segments or regions. Instead of searching the entire network configuration space, the scheduler segments the search into manageable portions, allowing parallel processing and reducing the overall scheduling time while working within the constraints of mesh/torus network topologies
Solution Approach 2:
The patent performs preliminary actions by pre-calculating and storing information about available resources and their configurations. This preprocessing allows the scheduler to make faster decisions during actual job allocation, reducing the time penalty associated with using mesh or torus network topologies
2Productivity
If parallel search process is executed to reduce scheduling time, then the productivity is improved, but the measurement precision deteriorates due to inaccurate detection of available resources
Solution Approach 1:
The patent implements feedback mechanisms where the parallel search processes communicate their findings and coordinate to ensure complete coverage of the search space. Results from parallel searches are aggregated and validated, ensuring that no available resources are missed while maintaining the speed benefits of parallel processing
Solution Approach 2:
The patent merges multiple parallel search results into a comprehensive view of available resources. By combining and consolidating findings from parallel search threads with proper synchronization, the system achieves both high productivity from parallelism and high measurement precision through complete resource detection
3Productivity
If the search range is divided into multiple threads for parallel processing, then the productivity is improved, but the device complexity increases due to inappropriate range allocation
Solution Approach 1:
The patent dynamically adjusts parameters such as search range size, number of parallel threads, and allocation strategies based on system state and job characteristics. This adaptability allows the system to optimize parallel search efficiency while managing complexity through parameter tuning rather than fixed complex structures
Data Source
AI summary
A device includes: a memory; and a processor coupled to the memory and configured to execute a process of managing data on a first subgraph that is included in a graph including vertices indicating computing resources of a system and edges indicating links between the computing resources and is provided for a first computing resource to which a first job are assigned, or data on a second subgraph that is included in the graph and connected to the first subgraph through a vertex indicating a computing resource to which none of the first job is assigned in the graph and that is provided for a second computing resource to which a second job is assigned, and a process of using the data to determine, based on the first subgraph, whether a third computing resource to which a third job is to be assigned exists.