Evaluating Service Mesh as a Network Monitoring Solution

Abstract: Modern software systems often utilize a distributed architecture, which uses the network as the communication channel. This results in the system being dependent on the network performance to behave as expected. The network performance varies, and network faults can occur, which may negatively affect the system. As a system maintainer, it is difficult to know if the problem lies in the service or the network. This thesis evaluates service mesh, an infrastructure layer that facilitates service-to-service communication, as a tool for network monitoring without having access to the network infrastructure. Experiments in a virtualized data-intensive environment running Istio and Kubernetes were performed, with the goal of evaluating how a service mesh aids in identifying network faults and its performance impact on the system. A latency increase and bandwidth decrease were discovered through the data provided by a service mesh. Packet loss, however, was not discovered. Using a service mesh resulted in a distinct negative performance impact on the system.