隨著各式各樣的物聯網服務崛起，現今採用集中式的傳統雲端運算已經不再能夠滿足各種需求，為了解決該問題，新型態的雲端運算架構-邊緣運算就被提出了，邊緣運算主要是將原先集中在骨幹網路雲端中心的運算伺服器改放置在邊緣網路，尤其是行動網路基地台中，利用運算伺服器靠近使用者的特點，來達成減少傳輸延遲的功效，也能讓骨幹網路的負載得以減輕。 邊緣運算伺服器與傳統雲端運算相比之下，邊緣運算伺服器運算能力並沒有這麼強大，再加上現今裝置有增無減的趨勢，期望邊緣運算系統能接受全部裝置的運算任務卸載要求(Offload Request)是不切實際的。然而其實許多裝置本身也擁有少部分的運算力，故邊緣運算系統是能夠適時地拒絕掉非必要的任務卸載要求，所以如何達成最高效的任務卸載決策與伺服器資源分配成為了在此領域中熱門研究的議題。 本論文提出Load-Adaptive Algorithm of Joint Resource Allocation(LAJRA)設計出自適應負載的方法，其在系統高負載時能將資源有效地保留給必要/危急的任務，而在低負載時又能接受非必要的任務卸載以減少資源的浪費，又因邊緣運算伺服器是架設在行動網路基地台中，故本方法特別整合上行頻寬與運算資源的分配期望相比其他方法能更符合實務面。 ;Along with the appearances of various IoT services, original centralized Cloud Computing hasn’t been able to satisfy every kinds of demands. For solving this problem, the new type of cloud computing structure -Edge Computing is presented. Its main concept is that setting up the computing server on edge network (especially base station) instead of centralized computing server in cloud center on backbone network, and utilizing the characteristic closing to UE to reduce the transmission delay and the load on backbone network. Comparing with traditional cloud computing with edge computing, the computing capacity of edge computing server is not powerful enough. Moreover, the number of mobile devices keep increasing, expecting that the edge computing system can accepts all offload requests of computing tasks is unrealistic. In fact, many devices have own small computing capacity so the edge computing system should be able to reject the offload requests of non-essential tasks. As noted above, the way how to achieve most efficient task-offloading decision and resource allocation is a popular research topic in this field. In this thesis, we propose Load-Adaptive Algorithm of Joint Resource Allocation(LAJRA) for adaptive load of edge computing system. The algorithm can reserve resources for the essential/critical tasks while the high system load, and it can also accept some offload requests of non-essential tasks for reducing the waste of resource while the low system load. Because the edge computing server is set up in base station, we integrate specially upload bandwidth and computing resource allocation in this algorithm to expect it will be more suitable in practice comparing with other methods.
