Dustdar's team unveils taxonomy for distributed continuum performance metrics

A team of nine researchers, including Praveen Kumar Donta, Boris Sedlak, and Schahram Dustdar from TU Wien, published a new paper on arXiv (2607.28407) proposing a comprehensive taxonomy of performance metrics for Distributed Computing Continuum Systems (DCCS). As AI and large-scale data applications increasingly span resource-intensive data centers and constrained edge devices, existing evaluation methods remain fragmented, focusing on isolated dimensions like computation or networking. This paper addresses that gap by categorizing metrics into three primary layers—computing-level, network-level, and application/user-level—while also introducing six emerging dimensions: sustainability, observability, adaptability, data locality, migration awareness, and continuum fragmentation.

Beyond the taxonomy, the paper provides mathematical formulations for each metric to support heterogeneous and dynamic continuum environments. It also outlines metric acquisition requirements, specifying whether a metric can be collected from a single node, multiple nodes, or the full system, and whether it suits operational monitoring or experimental evaluation. This structured approach enables developers and operators to consistently compare and improve cross-layer DCCS behavior, moving beyond siloed performance checks. The authors argue that as continuum architectures evolve, transparent and standardized evaluation becomes critical—both for optimizing system design and for benchmarking new algorithms. Practical implications include better tuning of edge-to-cloud pipelines, improved energy efficiency, and clearer trade-offs between latency, accuracy, and resource usage in AI workloads.