Adaptive Management of Microservices in Dynamic Computing Environments: A Taxonomy and Future Directions
A comprehensive taxonomy reveals that production dynamics are often only partially modeled in adaptive management systems.
A team of researchers led by Ming Chen and including cloud computing pioneer Rajkumar Buyya has released a comprehensive survey on adaptive management for microservices in dynamic computing environments. The paper, published on arXiv, addresses the challenge of managing microservice-based cloud applications that face changing workloads, evolving request paths, variable network conditions, interference, and failures. These dynamics require coordinated autoscaling, placement, routing, isolation, and remediation strategies.
The survey introduces a detailed taxonomy covering control locus, modeled dynamics, adaptation strategy, and evaluation evidence, with objectives and telemetry treated as cross-cutting concerns. By analyzing 84 system implementations and 13 evaluation artifacts, the researchers found that production dynamics are often only partially modeled in existing adaptive management systems. The reported performance gains are also highly dependent on the fidelity of the evaluation environment. The paper outlines key future directions including cross-layer coordination, telemetry-to-control abstractions, safe learning-based control, and reproducible dynamic evaluation to address these gaps.
- Taxonomy covers control locus, modeled dynamics, adaptation strategy, and evaluation evidence
- Analysis of 84 systems shows production dynamics are only partially modeled in most adaptive management approaches
- Reported gains depend heavily on evaluation fidelity; future work needed on cross-layer coordination and safe learning-based control
Why It Matters
This survey provides a roadmap for building more resilient cloud-native applications that can adapt to real-world, dynamic conditions.