Server Workload Prediction via Pattern-Adaptive Ensemble Modeling
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing ensemble prediction methods for server workloads rely on combining single prediction models for linear and non-linear characteristics, resulting in prediction values that are averages and fail to derive an optimum value.
Innovation Solution
A prediction device that discriminates between linear, non-linear, and mixed patterns in workload variation, selecting appropriate prediction models for each pattern to perform ensemble predictions, combining models with adjusted weights to improve accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If ensemble prediction combines single prediction models for linear and non-linear characteristics, then the prediction method can handle both types of patterns, but the prediction value becomes an average that fails to derive an optimum value
Solution Approach 1:
The patent segments the prediction process by first classifying the workload variation pattern into linear, non-linear, or mixed types, then selecting different prediction models accordingly. This segmentation allows the system to avoid averaging effects and derive optimum prediction values by using appropriate models for each pattern type.
Solution Approach 2:
The patent changes the parameter of model selection based on the detected variation pattern. When a linear pattern is detected, linear prediction models are selected; when non-linear patterns are detected, non-linear prediction models are selected. This dynamic parameter change optimizes prediction accuracy for each specific pattern type.
2Device complexity
If a single prediction model is used for all workload patterns, then the prediction method is simple, but it cannot accurately predict workloads with varying linear and non-linear characteristics
Solution Approach 1:
The patent introduces dynamics into the prediction system by making the model selection adaptive to the detected variation pattern. The system dynamically switches between different prediction models based on whether the workload exhibits linear, non-linear, or mixed characteristics, thereby improving accuracy without requiring a completely complex system architecture.
Solution Approach 2:
The patent performs preliminary classification of the variation pattern before selecting the prediction model. This preliminary action of identifying whether the pattern is linear or non-linear allows the system to prepare and select the appropriate model in advance, improving prediction accuracy while maintaining a structured and manageable approach.
3Adaptability or versatility
If ensemble prediction always combines linear and non-linear prediction models, then the prediction system can handle diverse patterns, but the prediction value depends on an average and cannot derive an optimum prediction value
Solution Approach 1:
The patent applies local quality by tailoring the prediction model selection to the specific local characteristics of the workload variation pattern. Instead of uniformly applying ensemble prediction, the system selects linear models for linear patterns, non-linear models for non-linear patterns, and combines them appropriately for mixed patterns, thereby deriving optimum prediction values for each local case.
Solution Approach 2:
The patent changes the composition parameters of the ensemble prediction based on the detected pattern type. For linear patterns, only linear models are selected; for non-linear patterns, only non-linear models are selected; for mixed patterns, both are selected in appropriate proportions. This parameter adaptation eliminates averaging effects and improves prediction precision.
Data Source
AI summary
A prediction device for predicting a workload of a server includes: a discrimination unit discriminating whether a variation pattern in the workload over time is only a linear pattern, only a non-linear pattern, or a linear/non-linear pattern in which the linear pattern and the non-linear pattern are mixed; and a prediction unit selecting a plurality of prediction models for linearity when the variation pattern is only the linear pattern, selecting a plurality of prediction models for non-linearity when the variation pattern is only the non-linear pattern, selecting the prediction model for linearity and the prediction model for non-linearity when the variation pattern is the linear/non-linear pattern, and deriving a future prediction value for the workload by performing an ensemble prediction combining the selected plurality of prediction models.