DevOps-oriented Operations for Microservices-based Service Composition over Cloud-native Infrastructure

Abstract

As cloud computing continues to grow recently, the new paradigm, cloud-native computing, is exploding in the cloud area. Accordingly, the existing monolithic architecture is evolving into a microservices architecture rapidly based on cloud-native computing. Moreover, by bridging the Internet of Things and the cloud together, various cloud applications are explosively emerging. These IoT-Cloud services require the formation of services that dynamically utilize geographically distributed cloud infrastructure, as IoT devices should be freely connected to adjacent clouds. Besides, we need the operational capability to provide an isolated environment for the agile and rapid cycle, including development and verification.

In this regard, the DevOps-oriented operational methodology of coordinating cloud-native infrastructure in a unified manner, taking into account multi-dimensional perspectives, is becoming important. Therefore, this dissertation presents a DevOps-oriented operation called Multi-X operations, supporting the MSA-based service composition over cloud-native infrastructure. Besides, to limit the scope of X in Multi-X operations to the range of sites, clusters, tenants, and workloads to the requirements of developers and operators, the following topics are addressed in detail to realize the proposed approach:

First, we propose a new type of cloud-enabled software concept called Dynamic OverCloud to address problems due to multiple clouds' usage for achieving an agile environment with the multi-sites operation. Dynamic OverCloud is a specially-arranged razor-thin overlay layer that provides users with an inter-operable and visibility-supported environment for MSA-based IoT-Cloud service composition over the existing multiple clouds. Then, we~design a software framework that dynamically builds the proposed concept. We also describe a detailed implementation of the software framework with workflows. To verify the concept, we apply a smart energy IoT-Cloud service case, an empirical example, by utilizing the software framework implemented on OpenStack cloud and Amazon AWS cloud infrastructure.

Second, to provide AI-feature supported cloud-native cluster with multi-tenants and multi-workloads aspects, we introduce a cloud-native intelligence cluster for flexibly supporting AI-inspired HPC(high performance computing)/HPDA(high performance data analytics) workloads. We design the detailed software framework for the proposed cluster. Then, we discuss various issues and approaches arising from building and operating the proposed cluster. To achieve multi-tenants, we also design the DevOps Portal. Finally, we perform the scenario that includes three AI workloads(parallelized training, distributed training, serverless inferencing) to identify our work's feasibility.

Third, to support multi-clusters aspects built using Kubernetes (K8S), a standard orchestration tool of cloud-native infrastructure, and to provide efficient service composition to developers, we propose microservices placement employing workload profiling. The proposed framework can identify and respond to workload characteristics in a practical way. To achieve this goal, we perform profiling experiments with selected workloads to derive delicate resource requirements. Then, we perform microservices placement derived from the profiled results. Finally, we show that the results of the experiments in applying our work are better performance.