Palette is a framework that uses distributed tracing datasets to generate representative microservice systems that can be used as macrobenchmarks by researchers for performing intervention experiments. Palette processes distributed traces to generate a system topology consisting of Graphical Causal Models (GCMs) for modeling the intrinsic system behavior. Palette then provides runtime bindings for converting the topology into an implementation enhanced with a GCM-enabled runtime that ensures that generated system behaves in a representative manner. Palette integrates these implementations with Blueprint to allow researchers to conduct a wide variety of intervention experiments across the vast design space of microservices.

Publications

Generating representative macrobenchmark microservice systems from distributed traces with Palette

Vaastav Anand, Matheus Stolet, Jonathan Mace, Antoine Kaufmann
Under submission, 2025.
[Preprint]