TU Berlin's new placement algorithm cuts serverless workflow costs by 10%

Serverless edge computing is a promising framework for running large workflows without users managing underlying infrastructure, but deploying those workflows across many heterogeneous servers and edge devices remains a challenge. The problem involves balancing monetary costs and workflow evaluation times while dealing with node-specific constraints. The team—Joshua Adamek, Natalie Carl, Trever Schirmer, Moritz Heinlein, David Bermbach, and Sergio Lucia—formalized this as a novel mathematical model: an integer nonlinear program (INLP) that captures both workflow dependencies and node attributes. Their paper, published on arXiv (2608.14427), directly tackles the complexity of finding optimal placements across large-scale cloud/edge fleets.

To make the model practical, the authors introduced a decomposition strategy that splits the placement problem into smaller subproblems, allowing it to scale to a larger number of nodes and handle decomposed knowledge of node attributes. In a case study, this approach showed strong scaling properties and delivered a mean improvement of 10% over a simple deployment heuristic. The work, submitted to IEEE for publication, is a significant step toward automating and optimizing serverless workflows in real-world distributed environments. For engineers and cloud architects, it offers a principled foundation for future placement algorithms that can cut costs and latency without requiring manual tuning.