Service Cluster Workload Prediction Using Cluster-Specific Models
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing workload prediction solutions for service clusters in cloud computing are limited in their ability to handle diversified online computing service demands and suffer from inaccurate prediction, leading to resource wastage or shortages.
Innovation Solution
A workload prediction method that involves obtaining load indicator sequences, determining service representations, performing clustering, and using pre-trained sequence prediction models specific to category clusters to improve prediction accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If a unified prediction model is used for all services, then the device complexity is reduced, but the prediction precision deteriorates because it cannot satisfy diversified service demands
Solution Approach 1:
The patent segments services into different categories based on their workload characteristics and prediction requirements. Multiple specialized prediction models are created for different service categories instead of using a single unified model. This segmentation allows each model to be optimized for its specific category, improving prediction precision while managing complexity through structured organization.
Solution Approach 2:
The patent applies local quality by creating prediction models with characteristics tailored to specific service categories. Each category receives a model optimized for its particular workload patterns and requirements, rather than applying a uniform approach across all services. This localized optimization improves prediction accuracy for each specific service type.
2Measurement precision
If multiple specialized prediction models are created for different service categories, then the prediction precision is improved, but the device complexity increases
Solution Approach 1:
The patent implements universality by creating a category classification mechanism that can handle multiple service types through a unified framework. The system uses a general category determination approach that works across diverse services, allowing multiple specialized models to be managed through a common interface and classification system, thereby controlling overall complexity.
Solution Approach 2:
The patent applies preliminary action by pre-defining service categories and their corresponding prediction models before actual workload prediction occurs. Services are classified into categories in advance, and appropriate pre-trained models are selected based on this classification. This preliminary organization reduces the complexity of model selection and management during runtime.
3Productivity
If historical workload data is used for prediction, then the prediction capability is enabled, but privacy information may be exposed
Solution Approach 1:
The patent extracts and removes sensitive privacy information from the historical workload data before using it for prediction. The system separates confidential details from the essential workload patterns needed for prediction, allowing prediction capabilities to function while protecting sensitive information from exposure.
Solution Approach 2:
The patent introduces an intermediary processing layer that handles historical data between the data source and the prediction model. This intermediary layer processes and anonymizes the data, acting as a buffer that enables prediction functionality while preventing direct exposure of sensitive information contained in the historical records.
Data Source
Figure 1
Figure 2
Figure 3~4
AI summary
Embodiments of this specification provide a workload prediction method and apparatus for a service in a service cluster. The method includes: obtaining a load indicator sequence of each service in a service cluster corresponding to a workload indicator in a same historical time period; determining, based on the load indicator sequence corresponding to each service, a service representation corresponding to each service; performing clustering processing based on the service representation corresponding to each service, to obtain a target category cluster to which each service belongs in multiple category clusters; obtaining multiple sequence prediction models pre-trained for multiple tasks, and enabling the multiple category clusters to correspond to the multiple tasks; and inputting at least a load indicator sequence of any service into a target sequence prediction model corresponding to a target category cluster of the service in the multiple sequence prediction models.