AIoT的發展，讓網路承受大量的負載，尤以雲集中式計算更造成網路延遲，這與AIoT往更嚴格的及時性功能的發展形成矛盾，而邊緣計算是目前最被看好的解決方式之一。因此我們研究一種符合AIoT異構網路環境與支持垂直水平分散計算的AI物聯網框架，使用多層架構、微服務以及虛擬化技術，讓框架內的各種服務可以在不同體系結構中快速遷移與運行，並透過人臉辨識專案的叢集閘道器範例，展示一個有效的邊緣計算系統。研究結果顯示，此運作方式與傳統開發方式、虛擬機施作方式相比，大幅減少開發與運行環境的建置次數，確保跨平台及落地執行一致的功能，期間也沒有產生明顯的額外開銷，其輕盈的映像檔，更大幅節省記憶體，可執行更多的副本，遷移速度也比較快，而且快速佈署與負載平衡機制，可以彈性對接不同的情境需求以及更多的服務請求。本研究成果可以提供開發者作為開發和運行經濟且高效的AIoT邊緣計算系統的參考平台。;The development of AIoT (AI+IoT) has caused the network to carry a lot of load, especially the centralized cloud computing, which causes even more serious network latency and conflicts with the development of AIoT’s stricter timeliness requirement, and the edge computing is currently as one of the most promising solutions. Therefore, we study an AIoT framework that conforms to the AIoT heterogeneous network environment and supports vertical horizontal decentralized computing. It uses multi-layer architecture, microservices and virtualization technology to enable various services within the framework to quickly move and execute in different architectures. We use the cluster gateway example of the face recognition project to demonstrate an effective edge computing system. Based on the research results of comparing with traditional development method and virtual machine method, this framework significantly reduces the number of building development and operation environments times, and ensure performing consistent functions when crossing platform and landing, meanwhile, it does not generate obviously additional overhead. The lightweight image which saves mush more memory, can be executed more much copies, shipped more faster, and the mechanism of quick deployment and load balancing can deal with different situation and more service requests flexibly. The results of this research can provide developers as a reference platform for developing and operating an economical and efficient AIoT edge computing system.
