Cloud Traffic Prediction with Graph-Structured API Traces
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Cloud applications generate substantial performance trace logs with nested API calls that are difficult to store and parse for accurate future API traffic prediction, hindering effective cloud resource planning and optimization.
Innovation Solution
A traffic prediction server retrieves and parses performance stack trace logs, storing them in a graph database to capture parent/child relationships, transforms the data into training data with spatial and temporal information, and uses a transformer model for accurate prediction.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If performance stack trace logs with nested API calls are stored as plain text, then storage is simple, but parsing difficulty increases and prediction accuracy decreases
Solution Approach 1:
The patent segments the complex nested API call structure into a hierarchical graph representation where each API call is a node and parent-child relationships are edges. This segmentation transforms the unstructured plain text into organized graph data that can be efficiently parsed and queried, resolving the contradiction between storage simplicity and parsing difficulty.
Solution Approach 2:
The patent introduces a graph database as an intermediary layer between the raw performance logs and the prediction model. This intermediary automatically parses and structures the nested API call data, eliminating manual parsing complexity while preserving the hierarchical relationships needed for accurate traffic prediction.
2Ease of operation
If nested API calls are stored in a relational database, then structured queries are enabled, but the tree-like semi-structured data structure becomes difficult to represent
Solution Approach 1:
The patent transitions from traditional two-dimensional relational database tables to a multi-dimensional graph structure that naturally represents hierarchical relationships. The graph database adds a new dimension of relationship modeling, allowing parent-child API call relationships to be represented directly as edges between nodes, thus enabling efficient queries without forcing semi-structured data into rigid relational schemas.
3Productivity
If cloud resources are manually adjusted to match traffic patterns, then resource optimization is achieved, but response time to traffic changes increases
Solution Approach 1:
The patent implements preliminary action by using the trained transformer model to predict future cloud traffic patterns before they actually occur. This allows cloud providers to proactively adjust resources in advance of predicted traffic spikes or drops, achieving both resource optimization and rapid response time by acting before the actual traffic change happens rather than reacting to it.
Solution Approach 2:
The patent establishes a feedback loop where actual traffic data continuously feeds into the prediction model, which then generates resource adjustment recommendations. This closed-loop system enables automatic, real-time resource optimization that responds immediately to changing traffic patterns, eliminating the delays associated with manual resource adjustment.
4Device complexity
If transformer model training data includes only basic traffic information, then data processing is simpler, but prediction accuracy decreases due to lack of spatial and temporal context
Solution Approach 1:
The patent transforms the training data by enriching it with spatial and temporal parameters extracted from the graph database. Each API call node is enhanced with temporal information (timestamps, duration) and spatial information (hierarchical position in the API call tree, relationship to parent/child calls). This parameter enrichment significantly improves prediction accuracy while the automated extraction process keeps data processing complexity manageable.
Data Source
AI summary
Some embodiments may be associated with a cloud computing environment. A computer processor of a traffic prediction server may retrieve performance stack trace logs from a traffic performance stack trace log repository that stores traffic information of the cloud computing environment. The traffic prediction server parses the performance stack trace logs as an objects list including parent/child object relationships and stores the parsed objects list in a graph database. The traffic prediction server may then transform the graph database into training data including spatial and temporal information and use the transformed training data to train a transformer model. According to some embodiments, the traffic prediction server also provides previous traffic input data to the transformer model when generates predicted traffic output data based on the previous traffic input data (e.g., to facilitate cloud load management).