Clustered Data Management in Virtual Docker Networks Spanning Geo-Redundant Data Centers : A Performance Evaluation Study of Docker Networking

Abstract: Software containers in general and Docker in particular is becoming more popular both in software development and deployment. Software containers are intended to be a lightweight virtualization that provides the isolation of virtual machines with a performance that is close to native. Docker does not only provide virtual isolation but also virtual networking to connect the isolated containers in the desired way. Many alternatives exist when it comes to the virtual networking provided by Docker such as Host, Macvlan, Bridge, and Overlay networks. Each of these networking solutions has its own advantages and disadvantages. One application that can be developed and deployed in software containers is data grid system. The purpose of this thesis is to measure the impact of various Docker networks on the performance of Oracle Coherence data grid system. Therefore, the performance metrics are measured and compared between native deployment and Docker built-in networking solutions. A scaled-down model of a data grid system is used along with benchmarking tools to measure the performance metrics. The obtained results show that changing the Docker networking has an impact on performance. In fact, some results suggested that some Docker networks can outperform native deployment. The conclusion of the thesis suggests that if performance is the only consideration, then Docker networks that showed high performance can be used. However, real applications require more aspects than performance such as security, availability, and simplicity. Therefore Docker network should be carefully selected based on the requirements of the application.