Distributed ML Model Execution With Data-Local Server Clusters
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Deploying machine-learning models in a production environment is challenged by the need to retrieve large quantities of historical data for training and execution, which consumes significant processing time and bandwidth, and existing systems struggle with incompatible model formats.
Innovation Solution
A method for distributed execution of machine-learning models on a server cluster, involving conversion to an executable format, execution on multiple nodes, and performance metric generation, allowing for real-time evaluation of models.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If large quantities of historical data are retrieved from storage for training and execution of machine-learning models, then model training and execution can be performed, but processing time and bandwidth consumption increase significantly
Solution Approach 1:
Instead of retrieving data from storage to the machine-learning model (traditional approach), the patent inverts the data flow by bringing the machine-learning model to the data storage location. The model is deployed to nodes distributed across the server cluster, allowing direct access to historical data at each node without centralized data retrieval, thereby eliminating bandwidth bottlenecks and reducing processing time while maintaining training accuracy
2Reliability
If large quantities of historical data are transmitted to the machine-learning model, then model training and execution can be performed, but bandwidth consumption increases significantly
Solution Approach 1:
The patent fundamentally inverts the traditional data processing architecture by deploying machine-learning models to distributed nodes where data resides, rather than centralizing data access. This eliminates the need for transmitting large quantities of historical data across the network, reducing bandwidth consumption to minimal model parameter updates while preserving training accuracy through distributed computation
3Device complexity
If machine-learning models are executed on a single computer, then model execution is simple, but processing speed and scalability are limited
Solution Approach 1:
The patent segments the machine-learning model into multiple distributed instances deployed across different nodes of a server cluster. Each node independently executes model operations on local data partitions, enabling parallel processing that dramatically increases execution speed and scalability while maintaining manageable system complexity through modular architecture
4Measurement precision
If multiple machine-learning models are tested to determine the best-performing model, then model selection accuracy improves, but development time increases
Solution Approach 1:
The patent enables parallel testing of multiple machine-learning models by distributing different model instances across multiple nodes simultaneously. Each node can independently evaluate different models or model configurations on the same or different data partitions, allowing comprehensive model comparison and selection to proceed in parallel rather than sequentially, thus maintaining high selection accuracy while reducing development time
Data Source
AI summary
Described are a system, method, and computer program product for distributed execution of a machine-learning model on a server cluster. The method includes initiating retrieval of a machine-learning model from a data repository and converting the machine-learning model to an executable format. The method includes transmitting the converted machine-learning model to each node of the server cluster and executing the converted machine-learning model on each node. The method includes generating an initial performance metric based on execution of the converted machine-learning model on each node. The method includes transmitting the plurality of initial performance metrics from each node to an external processor and combining the plurality of initial performance metrics to produce a combined performance metric. The method includes modifying a model hyperparameter of the machine-learning model based on the combined performance metric and executing the modified machine-learning model in a computer system to evaluate real-time event data.