| Summary: | 碩士 === 國立臺灣大學 === 資訊管理學研究所 === 107 === Nowadays, Internet-of-Things (IoT) sensors required more complicated computation power to serve video surveillance functions. With the advancement of edge computing technology and the birth of Artificial Intelligence chips make it possible for high computing capacity required services to be served.
Edge computing, also known as cloudlet, adopts the concept of edge cloud. When the service that requires high computing cost for IoT sensors to compute, the computation task will send to cloudlet to perform the task. Unlike traditional cloud computing, the sensors don''t have to send the raw data that route through multiple hops to the cloud for doing computation. In edge computing scenario, the sensors only need to submit the raw data to the edge cloud, which is the closest cloud to the IoT network, which significantly reduces the latency of the service.
However, due to the properties of IoT applications, the system tends to be inefficient and is not acceptable by the users as the following reasons. Such as insufficient computation power of sensor''s hardware, insufficient bandwidth between the sensors and the cloudlet and tolerable delay of real-time service.
Moreover, the computing capacity and detecting functions of sensors varied among sensor hardware. For example, when an event is detected, the coordinate sensor happens to unable to handle the event. This sensor needs to send the video to a sensor, cloudlet, or core cloud where it can handle the event. Moreover, our approach would use the constraint to make the offloading decision.
In our work, we focus on Smart City scenario. Our method not only can decide the offloading and routing strategy for each video gathering capability nodes but also considering the overall system constraints to maximize the minimum delay gap between the tolerable delay and the system service time. With this, the quality-of-experience (QoE) of the IoT system can be fulfilled.
|
|---|
