Non-Intrusive Cloud Application Transaction Pattern Discovery

Cloud application performance profiling is a complex task in view of multiple distributed micro-services running over varied physical and virtual machines, each implemented by different developers using diverse software components. Existing methods, usually as part of an cloud based Application Performance Management (APM) employ intrusive instrumentation techniques in order to map the calls between services depending on specific programming languages and affecting performance. We present Fibratus, an automatic non-intrusive solution for application service call pattern discovery of cloud applications. Fibratus is capable of discerning patterns of inter-service messages per transaction by employing modified Machine Learning clustering algorithms on live network traces. Fibratus can correlate service-specific protocol data patterns with transactional flow patterns to provide additional insights for performance profiling. Over an extensive set of experiments in the cloud, with various applications, utilizing different technologies and usage patterns, Fibratus demonstrated high accuracy and performance on interleaved traffic. In addition, Fibratus allows pluggable descriptive pattern discovery as demonstrated in our tests for example in SQL query, HTTP url and Redis REST calls.