Dominant Sequence Generation from Network Logs
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The sheer volume of logged information from networked computing devices makes it difficult to parse and mine for useful patterns or sequences of events, which could identify failure vectors and improve operational efficiency.
Innovation Solution
Artificially creating dominant sequences by partitioning logged information, generating a graph with weighted edges, and traversing it randomly to generate candidate trajectories, which are then filtered and scored for regularity and length.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of information
If logged information is retained in full volume, then completeness of event data is improved, but difficulty of parsing and mining for patterns increases
Solution Approach 1:
The patent segments the large volume of logged information into smaller, manageable partitions based on selected attributes. This segmentation allows the system to process and analyze specific subsets of logs independently, reducing the overall complexity of pattern mining while maintaining the ability to discover system-wide patterns through aggregation of partition-level insights.
Solution Approach 2:
The patent introduces an intermediary processing layer that transforms raw logged events into structured representations with associated metadata and hierarchical relationships. This intermediary structure serves as a bridge between the complete raw logs and the pattern mining process, making the data more amenable to analysis without losing information.
2Measurement precision
If all logged events are analyzed in detail, then detection accuracy of failure vectors is improved, but processing time increases
Solution Approach 1:
The patent applies partial action by initially analyzing a representative subset of logged events to identify common patterns and failure vectors. Rather than processing every single log entry with equal depth, the system performs targeted analysis on partitions that show anomalous behavior or higher relevance, achieving sufficient detection accuracy while reducing overall processing time.
Solution Approach 2:
The patent performs preliminary filtering and aggregation of logged events before detailed analysis. By pre-processing the log data to identify promising candidates for failure vector detection and organizing events into structured partitions, the system reduces the scope of subsequent detailed analysis, thereby maintaining detection accuracy while significantly reducing processing time.
3Loss of information
If graph traversal is performed exhaustively to generate all candidate trajectories, then completeness of sequence discovery is improved, but computational complexity increases
Solution Approach 1:
The patent performs partial graph traversal by generating a representative sample of candidate trajectories rather than exhaustively exploring all possible paths. This approach discovers the most significant event sequences while avoiding the computational burden of complete enumeration, particularly for large graphs with many nodes and edges.
Solution Approach 2:
The patent performs preliminary filtering of the graph structure before traversal by identifying and prioritizing important nodes and edges based on frequency and relevance metrics. This pre-processing step reduces the effective search space for trajectory generation, maintaining completeness of discovery for significant patterns while reducing computational complexity through focused exploration.
Data Source
AI summary
Dominant sequences that are representative of logged events can be artificially created. Initially, a graph comprising multiple nodes and edges between pairs of nodes is generated from logged information. The weights, or values, associated with edges are incremented as the log data reveals a temporal relationship between two nodes. Subsequently, a set of candidate trajectories, with each candidate trajectory representing a sequence of events, are generated by repeatedly traversing the generated graph in a random manner by commencing at randomly selected nodes and then proceeding in a random manner to subsequent nodes in accordance with the edge values, for a random quantity of steps. The candidate trajectories are filtered to eliminate those that are impossible or improbable based on a comparison between individual candidate trajectories and the quantity of occurrences within the logs. Scoring is based both on a quantity of occurrences as well as a quantity of steps.