Abstract:
Today, the idea of eliminating all non-coding responsibilities of developers has driven innovation in new methods, models and services. In this case, serverless computing appears as a new type of cloud service paradigm. Since AWS Lambda launched at the end of 2014, people pay more and more attention to serverless computing. This thesis explores the current state of serverless, including definition, structure, characteristics and performance. For enterprises and developers, the most direct benefits of urging them to use serverless have three points: reduced costs, automatic on-demand scaling, and no server maintenance. Serverless functions are automatically scaled and billed only for the time the code is running. However, serverless still has some shortcomings: it is stateless and has limit execution time; scaling to zero means the first time to start instances will encounter cold starts; serverless services also have I/O Bottlenecks which affect its performance. Through some benchmark tests mainly use AWS Lambda, I explored the performance of serverless function which is acceptable for latency-insensitive applications. Moreover, I designed a face recognition smart store recommender system based on serverless architecture. The system is a novel solution to the smart city concept, and the goal is to realise an automatic scaling recommender system for physical stores without the need to manage servers. Face recognition technology is used for identity authentication and activity recording of customers in physical stores, and then personalized recommendations are given by analysing user behaviour. At the same time, all data is stored in the cloud provider’s storage service, and all functions or interactions use serverless functions and third-party APIs. The system transplants specific advantages of online shopping to offline, and provides physical stores with the ability of personalized product recommendations and intelligent analysis of massive data. Although the effect of recommended products is lacking, the data problem is expected to be resolved in the future.