Executive Summary : | Edge computing is a technique that delivers low latency services by bringing compute resources closer to the end device. However, edge devices are only accessible over a limited range of distance, making it challenging to place services on them in the presence of user mobility. This project aims to address this issue by focusing on tail latency, which is often considered higher than 95 percentiles in wireless networks and edge devices. The project aims to optimize latency by introducing controlled redundancy, utilizing limited duplication of task execution on multiple edge devices, and leveraging multiple network paths for limited duplicate transmission. The multipath technique is proposed to provide more reliable transmission with a higher median time. The final goal is to place services on edge devices to minimize overall service latency considering user mobility due to the limited range of edge servers. The problems will be solved using reinforcement learning to learn the characteristics of edge devices and network connectivity while minimizing service latency. A miniature testbed of around 30 edge and user devices will be designed and a measurement study will be conducted to balance both median and tail latencies. The trade-offs involved in utilizing these techniques and optimizing latency will be studied as constrained optimization problems. |