Modular Chassis Failure Prediction via Edge Training Offload
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for training failure prediction models at a hardware level in computing environments consume excessive computing resources, leading to inefficiencies and resource bottlenecks.
Innovation Solution
A modular chassis manager offloads the training of failure prediction models across a network of edge devices, utilizing an offloading policy to distribute the workload based on system topology and resource availability, thereby reducing the computational burden on the manager.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If the modular chassis manager trains failure prediction models locally at hardware level, then model accuracy can be maintained, but computing resources are consumed excessively
Solution Approach 1:
The patent segments the model training process by dividing it into data collection phase (at edge devices) and model generation phase (at modular chassis manager). This segmentation allows accurate failure prediction while reducing the computational burden on the manager by having edge devices perform local data aggregation and preliminary processing.
Solution Approach 2:
The patent introduces an intermediary approach where the modular chassis manager acts as a coordinator that receives data from multiple edge devices, processes it centrally, and generates models that are then distributed back to edge devices. This intermediary role optimizes resource utilization by centralizing computationally intensive tasks while maintaining distributed data collection.
2Device complexity
If the modular chassis manager performs all training tasks locally, then model generation is simplified, but the manager becomes a bottleneck and cannot perform other tasks
Solution Approach 1:
The patent segments training responsibilities by having edge devices collect and pre-process data locally, while the modular chassis manager focuses on model generation and distribution. This division of labor enables the manager to perform other management tasks simultaneously, improving overall system productivity.
Solution Approach 2:
The patent implements continuous model generation and distribution cycles where the modular chassis manager can simultaneously coordinate data collection from multiple edge devices, generate models, and distribute them. This continuous operation allows the manager to maintain high productivity by overlapping multiple tasks in the training pipeline.
3Reliability
If training data is collected from all edge devices, then model accuracy improves, but data transmission and processing time increases
Solution Approach 1:
The patent segments data processing by having edge devices perform local data aggregation and preliminary filtering before transmitting to the modular chassis manager. This segmentation reduces the volume of data that needs to be transmitted and processed centrally, thereby reducing processing time while maintaining model accuracy through distributed data collection from all edge devices.
Data Source
AI summary
A modular chassis manager that implements a trained model for hardware prediction failure is programmed to: initiate a training of a machine learning algorithm; and in response to the initiating: make a first determination, based on an offloading policy, that the training is to be offloaded to one of the plurality of edge devices, wherein the offloading policy is generated by the modular chassis manager using system topology information obtained from the edge network, based on the first determination, select an edge device of the plurality of edge devices using the offloading policy, transfer training data to the edge device, instruct the edge device to perform the training using the training data to generate a trained model, and obtain the trained model from the edge device.