arXiv:2609.28943v1 Announce Type: new Abstract: The microservice architecture has been applied broadly in many mainstream computing environments. As one of the most prevalent environments, edge computing also widely employs microservices to handle diverse requests and tasks. In practical scenarios, microservices usually constitute workflows that are built based on service dependencies to execute
arXiv:2609.28943v1, titled "Dependency- and Layer-Aware Microservice Workflow Offloading and Service Image Caching for Edge Environments," was submitted on September 24, 2026, by Zhongxiao Wang, Yueshen Xu, and colleagues. The paper addresses the joint optimization of microservice workflow offloading and service image caching in edge computing to minimize task completion time.
The authors propose the DLA-HDRL framework, which utilizes Layer-Aware Cross-Attention (LACA) to exploit image layer sharing and Dependency-Aware Multi-Head Cross-Attention (DAMH-CA) to manage long-range task dependencies. This approach decouples the solution space using Hierarchical Deep Reinforcement Learning (HDRL) to handle the complexity of integer nonlinear programming.
Experimental results on a real-world DockerHub dataset demonstrate that the proposed method achieves a 22.38% reduction in average task completion time and significantly increases image hit rates compared to baseline algorithms like PPO and SAC. The study also contributes a new public dataset of microservice image layer configurations and open-source code.
The material addresses orchestration of microservice workflows in edge computing, where services are not independent tasks but are arranged into dependency-driven workflows. It focuses on deciding where to execute individual microservices—across edge nodes and, potentially, higher-tier infrastructure—while also deciding which service images should be cached locally. This is important because edge environments are resource-constrained, bandwidth-limited, and often heterogeneous, so naive per-service placement can create cold-start delays, redundant image transfers, and inefficient use of edge resources.
Its key contribution is the joint treatment of workflow offloading and service-image caching with explicit attention to dependencies and workflow layers. Rather than optimizing placement and caching separately, the work recognizes that these decisions are coupled: the set of images worth caching depends on where workflow stages will run, and the feasible, low-latency placement of a service depends on whether its image is already available at a node. By accounting for service dependencies and layered workflow structure, the approach can keep related services co-located or appropriately ordered, reduce unnecessary image provisioning, and avoid offloading choices that increase cross-node communication or violate execution precedence.
This matters because microservice-based edge systems increasingly rely on containerized or image-based deployment, where image availability can dominate startup time and end-to-end workflow performance. A dependency- and layer-aware strategy can lower workflow completion latency, reduce image download and setup overhead, and improve resource utilization across edge nodes. The work is therefore relevant to latency-sensitive, multi-tenant, and containerized edge applications where workflow structure, service reuse, and image provisioning are central to system efficiency.