邊緣運算（Edge Computing）是在網路的邊緣節點部署微型資料中心就近提供服務，利用邊緣網路的頻寬同時減少網路延遲。隨著物聯網（Internet of Things）技術快速演進，物聯網設備數量迅速增長，傳輸的網路流量劇增，因此物聯網服務提供模式勢必結合邊緣運算。具有移動性的物聯網設備會漫遊於不同場域間，使各場域服務請求的數量與種類不斷變化，因此微型資料中心的服務部署配置需要隨環境變化動態地調整，否則會使服務品質降低。面對突發的大量服務請求或是不支援的服務請求時，微型資料中心為了快速緩解狀況需要採取服務請求分流機制。然而缺乏其他微型資料中心的狀態資訊會導致不適當的分流決策。 本論文提出的Greedy-based Resource Orchestration for IoT Edge Computing （GROTEC）機制作用在邊緣運算環境下，透過全域的資源協調中心以貪婪策略調度微型資料中心的資源。GROTEC機制包含服務請求分派演算法、服務請求分流機制與服務遷移機制。系統實作上使用容器化技術降低部署服務的成本，並採用Kubernetes管理系統。為了降低物聯網設備的傳輸成本與提高資料中心處理訊息的效率，分別採用了 MQTT 協定與 Apache Kafka 訊息系統。實驗結果顯示GROTEC之服務請求分派演算法效能勝於基因演算法且只花費極少的計算時間；服務請求分流機制可迅速緩解超載狀態，並由物聯網閘道器分流服務請求可降低網路延遲並減少微型資料中心的網路流量；服務遷移機制可使雲端資料中心的負載下降，並且改善物聯網閘道器發送服務請求的網路延遲。 ;Edge computing is to deploy micro data centers (MDCs) at the edge nodes of the network to provide services nearby, using the bandwidth of the edge network while reducing network delay. With the rapid evolution of the Internet of Things (IoT) technology, the number of IoT devices has grown rapidly and the network traffic has increased dramatically. The IoT service delivery model is bound to incorporate edge computing. IoT devices with mobility roam around different areas, which makes the number and types of service requests in each area constantly change. The service resource configuration needs dynamical adjustment according to the environment state, otherwise the quality of services would decrease. When the MDC receives a sudden large number of requests or unsupported requests, the MDC needs to adopt a request offloading mechanism to quickly ease the situation. However, the lack of status information of other MDCs can lead to inappropriate request offloading decision. This paper proposes the Greedy-based Resource Orchestration for IoT Edge Computing (GROTEC) mechanism. The GROTEC mechanism uses a global resource orchestrator to orchestrate the resources of the MDCs with a greedy strategy. The GROTEC mechanism includes a request dispatching algorithm, a request offloading mechanism and a service migration mechanism. In system implementation, the GROTEC applies containerization to reduce the cost of service deployment, and the Kubernetes management system is adopted. Considering the transmission cost of the IoT devices and the efficiency of data centers, the MQTT protocol and the Apache Kafka messaging system are adopted respectively. The experimental results show that the request dispatching algorithm of GROTEC performs better than the genetic algorithm and cost very little computation time; the request offloading mechanism can quickly ease the overloading status, and the requests are offloaded by the IoT gateway to reduce network delay and reduce network traffic in MDCs; the service migration mechanism can reduce the load on the cloud data center and improve network delay for request transmissions from IoT gateways.
