Auto-constructing dataflow models from system execution traces

Abstract

This paper presents a method and tool named the Dataflow Model Auto-Constructor (DMAC). DMAC uses frequent-sequence mining and Dempster-Shafer theory to mine a system execution trace and reconstruct its corresponding dataflow model. Distributed system testers then use the resultant dataflow model to analyze performance properties (e.g., end-to-end response time, throughput, and service time) captured in the system execution trace. Results from applying DMAC to different case studies show that DMAC can reconstruct dataflow models that cover at most 94% of the events in the original system execution trace. Likewise, more than 2 sources of evidence are needed to reconstruct dataflow models for systems with multiple execution contexts.