HPC Resource Catalog Dimensionality Reduction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
High-performance computing (HPC) resources are difficult to organize and schedule due to their multidimensional nature, making it challenging to visualize resource availability for reservation and scheduling, which is an NP-complete problem.
Innovation Solution
A scheduling system that transforms multi-dimensional HPC resources into a one-dimensional schedulable resource catalog using a dependent graph structure, enabling predictable provisioning and providing visibility and multiple communication modes for scheduling and communication between entities.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If HPC resources are organized in multi-dimensional structure, then resource availability and scheduling capability are improved, but system complexity and difficulty of visualization increase
Solution Approach 1:
The patent transforms multi-dimensional HPC resources into a one-dimensional resource catalog organized by time slots. This dimensionality reduction simplifies the complex multi-dimensional resource structure (computing nodes, storage, networking, 3D locations) into a manageable 1D timeline view, making resource availability visualizable and schedulable while preserving the essential scheduling capability
2Ease of operation
If HPC resources are transformed into one-dimensional resource catalog, then scheduling and provisioning become simpler, but resource combination flexibility may be reduced
Solution Approach 1:
The patent pre-processes and transforms the complex multi-dimensional resource information into a simplified one-dimensional resource catalog before scheduling operations. This preliminary transformation organizes all resource combinations and their availability into a time-based structure, making subsequent scheduling and provisioning operations simpler while the pre-computed catalog preserves all necessary resource combination information
Data Source
AI summary
A scheduling system and method for high-performance computing (HPC) applications includes a network management component stored in physical memory and executed by a processor. The management component is configured to transform HPC resources into a schedulable resource catalog by transforming multi-dimensional HPC resources into a one dimension versus time resource catalog with a dependent graph structure between resources such that HPC resources are enabled to be provisioned into a service environment with predictable provisioning using the resource catalog. A graphical user interface component is coupled to the network management component and configured to provide scheduling visibility to entities and to enable a plurality of different communication modes for scheduling and communication between entities.