近年來，隨著雲端運算的普及，Serverless 架構因其高效、彈性的特性被越來越多公司所接納和廣泛採用。Serverless 通常會跟 Function as a Service (FaaS) 劃上等號，開發者只要專注在以函數 (Function) 為單位的業務邏輯開發而不用花費多餘的心力管理底層的伺服器，很多網路公司嘗試在自己的私有雲中架設 Serverless 平台，以往 Serverless 平台的架構都是以雲端供應商的角度設計，在自架的 Serverless 平台中有著諸多的限制，包含 Master side schedule pattern 問題，以及多數 Serverless 平台為了降低延遲預設平台內部的流量都透過未加密的 TCP 協定進行傳輸。如何改善現有的 Serverless 架構讓其更適合私有雲的場景是一個很重要的議題。 本論文為了改善 Serverless 平台的安全性以及 Gateway 的性能，導入 Sidecar based Function Instance 作為執行函數的運算實體，並且透過基於 Mutual TLS (mTLS) 加密的 gRPC 協定實現 Function Instance 之間的直接通訊，改善了 Master side schedule pattern 每個 Function invocation 都必須經過 Gateway 的缺陷。本研究基於 sidecar 架構開發的 Serverless 平台在默認加密的基礎上可以在多數場景比原先僅使用 TCP 傳輸的 Serverless 平台獲得更好的性能。本論文規劃了多個場景，將相同的函數部屬在本論文開發的 Serverless 平台以及目前最受歡迎的開源 Serverless 平台 - OpenFaaS，在不同的場景分析延遲，記憶體使用率等指標。實驗結果顯示本論文基於 sidecar 架構的 Serverless 平台開發的 API Gateway 相較於 OpenFaaS Gateway 能降低 100 % 的額外記憶體消耗，並使 API Gateway 的處理請求效率達到 OpenFaaS Gateway 的 2.688 倍。在延遲部份相較 OpenFaaS 平台，本論文開發的 Serverless 平台在函數鏈長度為 4 的測試中降低了 47.75 % 的延遲，並且 QPS=10000 的壓力測試中降低了 64.95% 的平均延遲。 ;In recent years, with the popularity of cloud computing, the Serverless architecture has been increasingly adopted and widely used by many companies due to its efficiency and flexibility. Serverless is often equated with Function as a Service (FaaS), where developers can focus on developing business logic in the form of functions without the need to manage underlying servers. Many software companies want to adopt the FaaS development model but do not want to be restricted to a specific cloud provider. Therefore, they choose to develop their own Serverless platforms.
Traditional Serverless platforms have been designed from the perspective of cloud providers, and self-hosted Serverless platforms have various limitations. These limitations include issues such as the Master side schedule pattern and the default use of unencrypted TCP protocol for internal traffic to reduce latency. Improving the existing Serverless architecture to make it more suitable for private cloud scenarios is an important issue.This study aims to improve the security of the Serverless platform and the performance of the gateway by introducing Sidecar-based Function Instances as the computational entities for executing functions. It also implements direct communication between Function Instances using the gRPC protocol with Mutual TLS (mTLS) encryption. This approach addresses the limitations of the Master side schedule pattern, where every function invocation must pass through the gateway. The Serverless platform developed in this study based on the sidecar architecture achieves better performance in most scenarios compared to the previous Serverless platforms that solely use TCP transmission.
