Evaluating the Performance of Docker Overlay Networks in High Latency Environments

The rise of Docker has popularized the deployment of microservices within application containers, facilitated by the introduction of libnetwork, which offers built-in support for connecting distributed containers through both native and thirdparty overlay drivers. However, overlay networks can introduce implicit overhead, and the geographic dispersion of containers may negatively impact performance. This paper investigates the performance characteristics of various Docker overlay solutions deployed in a high latency environment, utilizing the GEANT Testbeds Service for practical evaluation. We assess the deployment feasibility of different overlay solutions, specifically focusing on the native Docker overlay driver as well as third-party options such as Flannel and Weave. Performance is measured through a synthetic point-to-point benchmark and a streaming media application benchmark. Our findings indicate that neither benchmark exhibits significant performance degradation concerning latency or jitter across the evaluated solutions. However, throughput measurements for both UDP and TCP demonstrate irregular behavior, warranting further investigation into their underlying causes. This study contributes to understanding the operational efficiency of Docker overlay networks in challenging network conditions, providing insights for developers and engineers working with containerized applications