Comparative Analysis of Server-Based and Serverless Service Performance on Google Cloud Platform (GCP) (Case Study: Machine Learning Model Deployment)

Abstract

Cloud infrastructure providers such as GCP provide various computing services to deploy applications such as machine learning models, namely server-based and serverless. However, the two services each have different characteristics and advantages so that this becomes a difficulty factor for users in choosing cloud services. This research was conducted to compare server-based and serverless services with the aim of knowing the best service resulting from the analysis of performance measurements, namely CPU and memory utilization, latency, pricing, and developer experiences. The application of machine learning models is carried out on Compute Engine and Vertex AI services and will be tested for performance through requests to endpoints 100 times using JMeter for 30 minutes. The findings show that Vertex AI performance is better than Compute Engine with CPU utilization of 0.10%, memory utilization of 0.94%, and latency of 17.34ms but the cost efficiency is owned by the Compute Engine.

Keywords

• gcp
• server-based
• serverless
• deployment
• machine learning