Real-Time QoS Monitoring and Anomaly Detection on Microservice-based Applications in Cloud-Edge Infrastructure

Abstract

Microservices have emerged as a new approach for developing and deploying cloud applications that require higher levels of agility, scale, and reliability. A microservice-based cloud application architecture advocates decomposition of monolithic application components into independent software components called "microservices". As the independent microservices can be developed, deployed, and updated independently of each other, it leads to complex run-time performance monitoring and management challenges. The deployment environment for microservices in multi-cloud environments is very complex as there are numerous components running in heterogeneous environments (VM/container) and communicating frequently with each other using REST-based/REST-less APIs. In some cases, multiple components can also be executed inside a VM/container making any failure or anomaly detection very complicated. It is necessary to monitor the performance variation of all the service components to detect any reason for failure. Microservice and container architecture allows to design loose-coupled services and run them in a lightweight runtime environment for more efficient scaling. Thus, container-based microservice deployment is now the standard model for hosting cloud applications across industries. Despite the strongest scalability characteristic of this model which opens the doors for further optimizations in both application structure and performance, such characteristic adds an additional level of complexity to monitoring application performance. Performance monitoring system can lead to severe application outages if it is not able to successfully and quickly detecting failures and localizing their causes. Machine learning-based techniques have been applied to detect anomalies in microservice-based cloud-based applications. The existing research works used different tracking algorithms to search the root cause if anomaly observed behaviour. However, linking the observed failures of an application with their root causes by the use of these techniques is still an open research problem. Osmotic computing is a new IoT application programming paradigm that's driven by the significant increase in resource capacity/capability at the network edge, along with support for data transfer protocols that enable such resources to interact more seamlessly with cloud-based services. Much of the difficulty in Quality of Service (QoS) and performance monitoring of IoT applications in an osmotic computing environment is due to the massive scale and heterogeneity (IoT + edge + cloud) of computing environments. To handle monitoring and anomaly detection of microservices in cloud and edge datacenters, this thesis presents multilateral research towards monitoring and anomaly detection on microservice-based applications performance in cloud-edge infrastructure. The key contributions of this thesis are as following: • It introduces a novel system, Multi-microservices Multi-virtualization Multi-cloud monitoring (M3) that provides a holistic approach to monitor the performance of microservice-based application stacks deployed across multiple cloud data centers. • A framework for Monitoring, Anomaly Detection, and Localization System (MADLS) which utilizes a simplified approach that depends on commonly available metrics offering a simplified deployment environment for the developer. • Developing a unified monitoring model for cloud-edge that provides an IoT application administrator with detailed QoS information related to microservices deployed across cloud and edge datacenters.