雲端運算之服務要穩定執行，且需要有負載平衡和故障轉移機制、根據流量調整執行服務數量、系統開發和管理分工要明確以及按照使用量收費，所以出現了 Serverless 概念。目前公有雲的 Serverless 已經有很多效能測試的研究，但私有雲的部份還很少。因此本論文使用市面上常見的開源 Serverless 框架，有 Knative、OpenFaaS 、 Nuclio 和 Fission 建構在 Kubernetes 私有雲平台執行，比較不同 Serverless 框架在資源使用方面和執行效能的差異。本研究測試項目主要有 Serverless 框架資源使用狀態，了解到哪些 Serverless 框架較輕量。另外當服務副本數量縮小到零個時會發生冷啟動造成執行服務延遲，因此透過每間隔一段時間從 client 連到服務上驗證冷啟動對執行效能的影響，以及比較不同 Serverless 框架在執行函數同步和非同步的效能。最後本論文說明不同使用情境可以選擇哪些適合的 Serverless 框架做為本論文的研究成果。;Cloud computing services require stable execution, load balancing, and fault tolerance mechanisms. They also need to adjust the number of running service based on traffic and have clear division of labor in system development and management. Additionally, they are billed based on usage. These requirements have led to the emergence of the Serverless concept. Currently, there have been many performance tests conducted on Serverless in public clouds, but there is still limited research on private clouds. Therefore, this paper uses the normal open source Serverless frameworks, including Knative, OpenFaaS, Nuclio and Fission, to deploy on a Kubernetes private cloud platform, in order to compare the differences between the resource usage utilization and execution performance for different Serverless frameworks. The main task of this research is to test the resource usage status of Serverless framework and understand which frameworks are more lightweight. Furthermore, when the number of service replicas scaled to zero, the cold starts can be occurred can result in the executing service deploy. Therefore, the effect of cold starts on execution performance is verified by periodically connecting from the client to the service, and the performance of different Serverless frameworks in executing functions synchronously and asynchronously is compared. Finally, this paper explains which Serverless frameworks are suitable for different usage scenarios as the research result.
