There is a potential performance issue in every container-based platform. Since this issue comes from the fundamental limitation of the container engine, any platforms which depend on containers such as Kubernetes, Apache OpenWhisk, etc, can suffer from it. It is important to understand the issue and relevant factors to maximize the performance because things do not work in a way we expect in the real scene. In this session, I will share what I learned from my 3 years of experience in the serverless world. I will describe the performance characteristics of container engines and serverless platform, and how they can induce a performance issue. Big cloud providers such as AWS are also dealing with the issue and I will introduce their approach and the reason why we cannot take the same approach for all platforms. And finally, I will explain how can we overcome the issue. I could improve the performance around 150 times using Apache OpenWhisk.
As a committer and a PMC member of Apache OpenWhisk, I think I can give insights to audiences about the performance characteristics and issues of the container-based platform and how to overcome it. Also, Naver is a number one web portal which handles around half of the search traffics in Korea. So we can give practical lesson learned to audiences. First, audiences will understand what does performance mean in the serverless world and how different it is from other traditional systems. Second, attendees will learn what is the performance characteristics of the Docker engine and how can it affect the overall performance of container-based platforms. Third, I will share how AWS Lambda deals with this issue and what I did to maximize the performance in Apache OpenWhisk. Finally, I will present the performance comparison benchmark results and few guides to apply the same approach to other systems. Attendees will learn about the performance 101 in the container-based platforms.
