Neural Network Layer Mapping to Computing Resources
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing solutions fail to effectively utilize the processing capability of remote computing resources, leading to idleness and workload imbalance in computing resource pools, particularly in processing tasks such as those requiring multiple GPUs for deep learning and machine learning applications.
Innovation Solution
A method that determines parameter data for neural network layers, ranks these layers, and matches them with computing resources based on status information to optimize resource allocation, ensuring that each layer is processed by the most suitable computing resource, thereby maximizing the utilization of available resources.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Power
If computing tasks are processed using remote computing resources in a resource pool, then the processing capability can be expanded, but idleness and workload imbalance arise in the resource pool
Solution Approach 1:
The patent segments the computing task into multiple sub-tasks corresponding to different layers of the neural network model. Each layer is independently assigned to a computing resource, enabling granular control and distribution of workload across the resource pool, thereby reducing idleness and balancing workload.
Solution Approach 2:
The patent dynamically adjusts the mapping between layers and computing resources based on real-time status information (such as workload, performance metrics) of the computing resources. This parameter-based dynamic allocation optimizes resource utilization and prevents both idleness and overload.
2Productivity
If multiple computing resources are used to process neural network layers, then processing throughput can be increased, but workload imbalance occurs across resources
Solution Approach 1:
The patent implements a feedback mechanism where status information of computing resources is continuously monitored and used to adjust the layer-to-resource mapping. This feedback loop ensures workload balance by reassigning layers from overloaded resources to underutilized ones, maintaining balanced workload distribution while preserving high throughput.
Solution Approach 2:
The mapping between neural network layers and computing resources is made dynamic rather than static. The system adapts the allocation in real-time based on changing resource status, enabling the system to maintain both high throughput and balanced workload distribution as conditions change.
3Ease of manufacture
If computing resources are allocated without optimization, then resource allocation is simple, but resource idleness and imbalance occur
Solution Approach 1:
The patent performs preliminary ranking of computing resources based on their status information before allocating layers. This pre-processing step organizes resources in an optimal order, making the subsequent allocation process systematic and efficient while maximizing resource utilization without excessive complexity.
Data Source
AI summary
A method for processing a computing task comprises: determining parameter data of multiple layers associated with a neural network model in response to receiving a computing task based on the neural network model; ranking at least one part of the multiple layers on the basis of the parameter data so as to obtain a layer list; ranking multiple computing resources on the basis of status information of the multiple computing resources so as to obtain a resource list; and determining a mapping between a corresponding layer among the multiple layers and a corresponding computing resource among the multiple computing resources on the basis of the layer list and the resource list, where the mapping indicates one computing resource among the multiple computing resources is to process parameters associated with one layer among the multiple layers.