物聯網平台的效能管理

Abstract

物聯網被視為是繼電腦、網際網路之後新一波的資訊革命。隨著智慧型手機和平板電腦的普及，它們儼然成為了物聯網裝置的操控中樞，各種不同領域的物聯網的應用也紛紛被創建出來。物聯網核心的商業模式都是圍繞著資料轉。這些透過各種物聯網裝置感測而來的數據會被收集起來加以分析提供物聯網應用進一步的使用。為了達成這個目的，需要整合物聯網結構中的的感測層、網路層以及應用層。物聯網共同服務平台的觀念正在慢慢成形以達到整合這三個架構的目的並提供應用層及感測層的開發者更有效率的開發各自的應用及服務。 一個理想的物聯網共同服務平台是要能夠讓各種不同的裝置都可以互相溝通並讓資料可以被有用的應用共享。在未來將有大量的物聯網應用和設備產生大量的數據，並產生繁重的工作負載給物聯網的網路及平台。因此，物聯網平台要有能夠應付瞬間湧進大量流量的壅塞機制及對這些資料做優先級分類。然而，到目前為止現有的研究很少有此效能管理的機制。 本論文提出了一個具備有效能管控機制的物聯網平台。它允許用戶在向物聯網平台註冊時設定其優先級。除此之外，平台可以透過壅塞管控機制來達到平台效能管理的能力。在我們提出的架構中，當有突發的大流量湧入系統時，節流調度機制就會被觸發並開始拒絕進入平台的請求從而維持平台的效能及穩定性。 為了比較我們提出的方法對原有架構的改進，我們設計了三個實驗來評估我們的架構。我們比較了突發性大流量之下兩個架構的效能表現、評估緩衝器大小對新架構的影響以及新架構對於多個不同優先權應用的效能表現。結果證實，我們提出的架構不僅可以有效地增進系統的效能更達到保證不同優先級應用的效能。

The Internet of Things (IoT) is regarded as a new revolution followed by the invention of computer and the Internet. With the proliferation of smart phones and tablets, they have become the main control center of the IoT devices to enable the creation of various IoT applications. Data is the core of the IoT business model. These data sensed by the IoT devices will be collected and analyzed for further usage by IoT applications. In order to make this happened, the integration between the sensor, network and application layers is required. The concept of IoT/M2M common service platform is thus emerging to accomplish this integration and enables the developers of both sensor and application layers to build their services and applications more easily. An ideal common service platform should have the ability to enable communications between devices and let the data be shared for useful applications. Thus in the future there will be massive IoT applications and devices producing large amount of data and generating heavy work load to the IoT/M2M network and platform. As a result, an IoT service platform should be capable of handling such burst load with congestion control and classified priority. However, until now there has been little research focusing on such performance management mechanisms. This thesis proposed a platform architecture that incorporates the long wanted performance management functionalities. It allows the users to define the priority of their applications when registering to the platform. Furthermore, it is capable of to managing the system performance via congestion control. In our proposed architecture, the throttling scheduler will be activated when there are bursting incoming requests. It will reject the incoming requests in order to maintain the stability and performance of the platform. To compare our proposed approach against the original one, we design three tests to evaluate our architecture. We compared the performance of two architectures under a constant and a burst load, evaluated the impact of the buffer length to the proposed approach and measured the performance of multiple priority applications in the proposed architecture. The results verified that our proposed architecture can not only effectively enhance the system performance but also guarantee the performance of applications under different priority levels.