Offload Advisory System - 行動雲端之效能監控與排程機制

Abstract

近年來智慧型手機普遍流行於大眾, 手機的處理效能和電池壽命是開發者目前遇到的巨大瓶頸。 如果能將這些工作傳送到雲端伺服器進行處理, 我們將能改善目前所遇到的困境。 對一個行動應用程式來說, 為了克服自身裝置運算的限制, 就需要更多的雲端資源。 流水型編譯模式的應用程式打破了這些限制, 並利用程式的可攜性在不同的行動雲端運算環境下得到分佈式的效能。 因此我們提出一個架構, 稱作行動雲端效能監控排程機制系統, 此系統可流水型編譯模式下的應用程式中的元件能充分利用雲端的資源。 我們將介紹所提出的系統如何整合之前實驗室開發的虛擬效能分析工具來找出熱點, 以及整合支持向量機到我們的決策管理者來做動態卸載決定。最後我們將討論我們架構的設計方式以及探討在不同的環境情況下我們得到的實驗數據。

In light of the growing popularity of smartphones, the processor speed and battery life on such device have been the major bottlenecks for advance mobile computing. If the compute-intensive tasks can be preformed by the cloud server efficiently, we could solve the bottlenecks. For a mobile application, to overcome the resource limitation on its own device, we propose to adopt the Flow-Based Programming (FBP) model to enable the development of portable mobile applications that run across different mobile-cloud computing environments with scalable performance. In this thesis, we integrate the efficiency of offloading task to the cloud and propose a mechanism, called Offload Advisory System (OAS), to guide the offload of components for good performance and proper resource utilization. OAS leverages our previous work, Virtual Performance Analysis (VPA), help locate the hotspots in Android applications, and uses extit{support vector machine} (SVM) into for making dynamic offloading decisions based on application profile and current input data. The thesis discusses the design of the proposed framework and presents several usage scenarios in our experimental studies.