基於霧運算之低延遲穿戴式服務虛擬化

Abstract

穿戴式裝置日漸普及並且具有許多方便的功能,例如:心跳頻率偵測、健身運動追蹤系統。儘管如此,由於穿戴式裝置必須非常輕便以及省電,所以其運算能力會受到相當大的限制。現存的解決方案是將穿戴式裝置與協調設備(Local-Hub)通常為手機,透過低耗電藍芽通訊協定進行配對。配對完成後,他們可以透過低耗電藍牙通訊協定或是 Wi-Fi 彼此進行溝通。預設來說,每當服務需求發生,穿戴式裝置會將其傳送至協調設備,並等待服務完成以及結果。如果他們是透過低耗電藍芽溝通,他們必須在同一個狹小的連線範圍之內。然而透過 Wi-Fi 溝通雖然可以不受此限制,但回應時間會因此延長許多。在這篇碩士論文內,我們實作了一套基於虛擬化協調設備(Virtual Local-hub)概念之系統去克服上述的缺點。我們在邊際網路裝置上佈建穿戴式服務,並修改穿戴式裝置的系統行為,使穿戴式裝置所需之服務在邊際網路即可完成,不需要進入再網際網路。更重要的,我們對於穿戴式裝置的系統修改是透明化的(Transparent),所以現存的應用程式可以自然的融入我們的系統,不需要被更改或重新撰寫。基於邊際網路裝置之運算能源是受限的,並不像雲端伺服器一樣充足,我們提供的穿戴式是共享且可以被遠端存取的,所以每個邊際網路裝置的運算量將可以達到平衡,不置於超載。最後我們設計了全面的實驗來評估我們提出的系統,實驗結果顯示出此系統可以大幅縮短應用服務的執行時間,最高至百分之六十,並且不會造成穿戴式裝置多餘的系統負擔。

Wearable devices have become ubiquitous and provide convenient features for users, such as heart rate monitor and fitness tracker. However, because the devices are designed to be light-weight and power-saving, the computing capability of wearable devices is often limited. The existing solution is to pair wearable devices with the local-hub, most are smartphones, via Bluetooth Low Energy (BLE). After pairing is completed, they can communicate with each other via BLE or Wi-Fi interface. By default, whenever service requests take place, wearable devices automatically transfer it to the local-hub and wait for the results. Communication via BLE, however, is only functional within a narrow connection range, whereas communication via Wi-Fi interface suffers from the long response time. In this thesis, we propose a system design based on the concept of Virtual Local-Hub (VLH) to address these shortcomings. The proposed system deploys wearable services at edge devices and modifies the system behavior of wearable devices. Consequently, wearable devices can be served at the edge of network without entering into the Internet. Most importantly, the system modifications on wearable devices are transparent, so the existing applications can fit into our system naturally. Considering the limited computing capacity of edge devices, the execution environment should be light-weight. So edge devices provide common and native function module as a shared wearable service. We conduct comprehensive experiments and the results show that the execution time of wearable services can be reduced by up to 60% with low system overhead.