COMPARATIVE ANALYSIS OF SERVICE MESH IMPLEMENTATION PERFORMANCE ON KUBERNETES

• Main Author: Xavier Rafifsyah Prasetyo, M
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/65839
• Institution: Institut Teknologi Bandung

Service mesh is a software to enhance the capability of a microservice applications network. Service mesh has many implementation and sometimes choosing which one to use is a difficult task. Service mesh performance needs to be tested and analysed so that the right service mesh can fit an operator needs. To evaluate a service mesh performance, a generated traffic load is directed to a running application installed with a service mesh. Latency and resource usage is then collected during and after the test. To do the evaluation efficiently, a script is created that will test the service mesh implementation one by one. Evaluation are conducted in a cloud environment. The resulting data shows that each implementation of service mesh has their own strength and weakness. The efficient service mesh are as follows, from the most efficient to the least efficient: Kuma, Consul, Linkerd, Nginx Service Mesh, and Open Service Mesh. Kuma has the best performance with an average latency of 2.18 milisecond, average proxy CPU usage of 460 miliCPU and average proxy memory usage of 6.3 GB. On the other hand, the highest recorded average latency is Istio with a value of 8 milisecond. Open Service Mesh has the highest proxy resource usage with an average CPU usage of 2000 miliCPU and average memory usage of 24.6 GB.