Research on A Lightweight Service Mesh Architecture Using eBPF

Many applications deployed in the cloud are typically refactored as widgets in microservices, and these components run as containers in a Kubernetes environment. The containers then run on clusters of physical servers that interoperate with the data center network. With this deployment, a large number of microservice containers running comes with resource consumption. The prevailing management architecture of microservices deepens this resource consumption on physical servers and increases network latency. We believe that this scenario could be further optimized. In this paper, we argue for an improved service mesh architecture design (compared to the traditional one). It reduces the resource consumption and improves the performance of deployed microservice applications, while providing some end-to-end visibility to debug problems that occur in microservices. We present the design and provide preliminary implementation details using eBPF technology to illustrate the feasibility of the system.