Service Cluster Workload Prediction with Attention Graphs
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing workload prediction methods for serverless architectures in service clusters suffer from low accuracy, particularly when relying on historical load information, which may include privacy data, and fail to optimize resource allocation effectively.
Innovation Solution
A method involving constructing similarity graphs based on correlation coefficients of system status and workload indicators, using an attention mechanism and convolutional networks to aggregate service representations, and performing graph and time convolutional processing to predict future workloads, incorporating system status and periodic patterns.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If workload prediction is based on historical load information of a single service, then the prediction process is simple, but the prediction accuracy is low
Solution Approach 1:
The patent merges multiple services' indicator sequences into a unified prediction model. By constructing similarity graphs that connect services based on their indicator correlations and using graph convolutional networks to aggregate information across the service cluster, the system combines data from multiple sources to improve prediction accuracy while managing complexity through structured aggregation
Solution Approach 2:
The patent introduces similarity graphs as an intermediary structure between raw indicator sequences and prediction outputs. The graph convolutional network processes these intermediate representations to capture service relationships, acting as a mediator that transforms individual service data into cluster-level predictions without directly processing all raw data
2Measurement precision
If multiple system status indicators are collected from multiple services, then the prediction accuracy can be improved, but the data processing complexity increases
Solution Approach 1:
The patent segments the complex prediction task into distinct processing stages: constructing similarity graphs from indicator sequences, performing graph convolutional processing to aggregate service representations, and finally predicting workload indicators. This segmentation allows each stage to handle specific aspects of the data independently, reducing overall processing complexity
Solution Approach 2:
The similarity graph structure serves multiple functions simultaneously: it represents service relationships, aggregates indicator information, and provides the basis for graph convolutional processing. This multi-functional structure reduces the need for separate processing mechanisms for each function, simplifying the overall system
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 n+1 indicator sequences respectively corresponding to services in a same historical time period, which include n status indicator sequences corresponding to n system status indicators and a load indicator sequence corresponding to a workload indicator; calculating a correlation coefficient of a same indicator between different services based on the n+1 indicator sequences corresponding to the services, and constructing a similarity graph corresponding to the same indicator, to obtain n+1 similarity graphs; processing the n+1 similarity graphs by using an attention mechanism, to obtain an aggregated similarity graph; performing, by using a first convolutional network, graph convolutional processing on an initial representation of each service in the aggregated similarity graph, to obtain an aggregated representation corresponding to each service; and performing workload prediction based on the aggregated representation.