Microservice Identification via Runtime Trace Clustering

The patent introduces runtime traces as an intermediary data structure that captures business contexts and temporal dependencies during application execution. These traces serve as a mediator between the monolithic application code and the microservice identification process, enabling accurate clustering by preserving contextual information that traditional static analysis methods miss.

The patent replaces traditional mechanical/static code analysis methods with a dynamic machine learning-based clustering approach. Instead of relying on fixed rules or simple dependency graphs, the system uses trained models that process runtime trace data to automatically identify microservice boundaries, substituting rigid mechanical processes with adaptive intelligent systems.

The patent performs preliminary actions by instrumenting the monolithic application code beforehand to automatically capture runtime traces during normal execution. This preprocessing step collects all necessary business context and temporal dependency information in advance, transforming complex raw execution data into structured trace formats that are easier for clustering algorithms to process later.

The patent creates copies of the monolithic application's runtime behavior in the form of trace data structures. Instead of directly analyzing the complex original execution flow, the system generates simplified copy representations (traces) that preserve essential business context and temporal relationships, making the data more manageable for subsequent clustering operations.