Joint SDN Controller and Microservice Placement for Scalable Deployments in Computing Continuum Networks

Abstract

The Internet of Things (IoT) paradigm is a crucial enabler for the computerization of real-world processes, and is becoming especially attractive in intensive domains. However, the critical IoT applications used in such environments require very high Quality of Service (QoS) and low economic deployment costs to be practical. To achieve the necessary QoS and costs, it is paramount to deploy the application in a distributed Cloud Continuum environment, by disaggregating it into a Microservices Architecture, and automating service discovery exploiting the Software-Defined Networking (SDN) paradigm. \new{In this context, the placement and replication of both SDN controllers and microservices greatly impacts the network QoS and, by extension, the overall application QoS, as well as the deployment costs. Managing the placement and replication of microservices and SDN controllers is especially complex to do manually, especially considering that microservice placement affects the QoS and cost provided by a given SDN controller placement and vice-versa. However, while there are initial approaches to address these problems, they are highly time and resource-consuming, limiting the support to large scenarios.} To tackle these issues, this work proposes the Microservice and SDN Controller Workflow Heuristic (MCWH), a multi-objective algorithmic heuristic to replicate and place SDN controllers and microservices in the Cloud Continuum efficiently and supporting large scenarios. The main novelty of MCWH is considering both microservice and SDN controller replication and placement not as separate problems, but as a single optimization effort that considers the effects across the problems, which so far only optimal methods could consider. The evaluation in a realistic healthcare use case shows that MCWH can achieve up to 70.48X speed-up w.r.t. optimal methods, with an optimality gap as low as 8.2%, allowing it to improve the QoS and cost in large scenarios, which need automated solutions the most.