AUTOSCALING OF MICROSERVICE BASED APPLICATION USING SERVICE DEPENDENCY ON KUBERNETES

• Main Author: Susanto, Yoel
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/55473
• Institution: Institut Teknologi Bandung

As software grows in complexity, developing software in a monolithic manner becomes harder because of the various overheads that grow linearly with the application size. To address such issue, software development shifts to create many smaller applications with their own specific functions. Such development methodology is known as microservices. Although microservice application adapt better to growing business needs, operating microservice application is relatively harder because the application is divided into many smaller parts. One such challenges is applying autoscaling effectively. The autoscaling has to be targeted at the correct service, otherwise the autoscaling process will result in an suboptimal performance and waste compute resources. In this research, we developed an autoscaling system that takes into account the interconnectivity between services in a microservice application. Service interconnectivity data is obtained by recording HTTP request calls in the cluster. The autoscaling system also incorporated historical CPU usage data to calculate the correct amount of replication and determine if a service is really relevant for being auto scaled. During our test using a linearly increasing load, we found out the system experienced 128% increase in throughput using our new autoscaler. This increase in performance is achieved while making sure of efficient usage of compute resource.