Performance Investigation Tool for Distributed Processing Systems
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Training deep neural networks (DNNs) is resource-intensive and time-consuming due to the need for processing large amounts of data and learning numerous parameters, which can be exacerbated by the complexity of distributed processing systems.
Innovation Solution
A computer-implemented performance investigation tool (PIT) is developed to assess the performance of distributed processing systems before deployment, allowing for the determination of time-based performance measures and optimal configuration of computing units for efficient graph processing tasks, including DNN training.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If a distributed processing system is used to train deep neural networks in parallel, then training speed and productivity are improved, but system complexity and resource coordination overhead increase
Solution Approach 1:
The patent applies preliminary action by performing performance investigation and prediction before actual DNN training deployment. The system evaluates computing units, predicts training performance, and optimizes configuration in advance, allowing the distributed processing system to be properly prepared before use, thereby reducing coordination overhead and improving training speed without escalating complexity during execution.
2Adaptability or versatility
If manual configuration of distributed processing systems is performed, then adaptability to specific tasks is improved, but time consumption and operational complexity increase
Solution Approach 1:
The patent implements feedback by using performance prediction results to automatically adjust and optimize the configuration of distributed processing systems. The system predicts training performance based on computing unit characteristics and DNN parameters, then uses this feedback to automatically select optimal configurations, reducing manual configuration time while maintaining high task adaptability through data-driven decision-making.
3Ease of operation
If deployment of distributed processing systems is performed without prior performance assessment, then ease of deployment is improved, but resource waste and training inefficiency increase
Solution Approach 1:
The patent applies preliminary action by conducting comprehensive performance investigation and prediction before deployment. The system evaluates computing unit performance characteristics, predicts DNN training performance metrics, and identifies optimal configurations in advance. This pre-deployment assessment ensures resources are properly allocated and configured, preventing resource waste and training inefficiency while maintaining ease of deployment through automated optimization.
Data Source
AI summary
A performance investigation tool (PIT) is described herein for investigating the performance of a distributed processing system (DPS). The PIT operates by first receiving input information that describes a graph processing task to be executed using a plurality of computing units. The PIT then determines, based on the input information, at least one time-based performance measure that describes the performance of a DPS that is capable of performing the graphical task. More specifically, the PIT can operate in a manual mode to explore the behavior of a specified DPS, or in an automatic mode to find an optimal DPS from within a search space of candidate DPSs. A configuration system may then be used to construct a selected DPS, using the plurality of computing units. In one case, the graph processing task involves training a deep neural network model having a plurality of layers.