We have investigated critical issues related to the design and performance of edge systems in heterogeneous environments. Edge computing has emerged as a promising paradigm for providing low-latency and high-bandwidth services at the network edge. We have produced the following major outcomes.

Data-centric wireless communication

Data sharing with multiple nodes on a peer basis at high rate, low loss is needed in many edge scenarios such as vehicle, IoT and drone applications. Existing wireless technologies do not offer this critically needed function. We developed a new wireless communication technology based on a publish/subscribe paradigm at radio level, diverging radically from the point to point paradigm in existing wireless technologies. This enabled us to create V-MAC, a data centric radio design that can share data with peer devices at high rate, low latency with one-to-many multicast capability. We developed multiple prototypes which have delivered this critically needed function at high performance. It will be an enabler for many edge based applications for vehicles, IoT and drones.

Task Offloading and Resource Allocation for Heterogeneous Edge Environments:

Task offloading has emerged as a potent solution to handle data-intensive tasks and mitigate network congestion in edge environments. However, it presents its own challenges, particularly concerning resource allocation in heterogeneous environments. We have addressed these challenges through offloading tasks and allocating resources jointly.

We formulated a joint task offloading, access point selection, and resource management problem. We decomposed it into two subproblems: the joint task offloading and computing resource allocation problem and the joint base station selecting and bandwidth allocation problem. We designed polynomial-time algorithms with provable approximation ratios for the two subproblems. In addition, we proposed a deep learning-assisted online algorithm that can make fast decisions.

Moreover, through extensive simulation and testing, we demonstrated that the proposed algorithms outperformed baselines and were near-optimal over a wide range of settings. The deep learning-assisted online algorithm is around 500× faster than the proposed approximation solver and is near-optimal. Therefore, our research contributes significantly to improving task offloading and resource allocation in heterogeneous edge environments.

Network Function Virtualization and Resource Allocation in Edge:

The integration of Network Function Virtualization (NFV) into edge computing opens new avenues for efficient network management. Our research primarily focused on creating an effective virtual network function (VNF) deployment and resource allocation mechanism to minimize overall latency in edge networks and designing a backup VNF backup scheme to ensure availability.

We first proposed a joint service function chain deployment and resource management problem considering the suitabilities between VNFs and physical servers in edge environments to minimize system latency and proposed a game-theoretic-based algorithm with a constant approximation ratio. In addition, to ensure the availability of vulnerable VNFs on edge servers, we proposed an online VNF backup problem under availability constraints, for which we proposed an online algorithm and proved that it has near-optimal performance.

Furthermore, real-world data-driven simulation results demonstrated that our joint service function chain deployment and resource management algorithm is time-efficient and the performance is much better than the baselines and close to the optimal solution. In addition, the proposed VNF backup scheme significantly outperforms popular baselines used in practice. Thus, our work significantly contributes to optimizing NFV placement, resource allocation, and availability guarantee in edge computing.

Distributed Edge Caching for Mobile Edge Networks:

Distributed and decentralized edge caching is a crucial aspect of optimizing mobile edge networks. With the increasing demand for low-latency and high-quality services, edge caching plays a significant role in reducing congestion and improving the overall user experience. In this project, we have focused on studying and developing efficient strategies for distributed and decentralized edge caching in mobile edge networks. One of our key contributions is the design of a distributed caching optimization problem considering the utilization of non-volatile memory (NVM) devices. We proposed and studied a two-layer storage system (NVM-SSD) in the edge network.

We formulated an edge caching problem to minimize the total serving delay of the network. We developed a distributed algorithm with a convergence guarantee, which allows full parallelization for large-scale networks. We further developed a fully decentralized algorithm for scenarios without any coordination of the remote cloud. The decentralized algorithm also has a convergence guarantee.

Our research also involved extensive trace-driven simulations and experiments in a small-scale system to evaluate the performance of distributed and decentralized edge caching. The results demonstrate the significant benefits of leveraging NVM in edge caching, as it significantly improves data offloading efficiency and reduces serving delay. The proposed algorithms outperform existing ones, highlighting the effectiveness of our approaches in optimizing edge caching in mobile edge networks.

Last Modified: 08/31/2023
Modified by: Yuanyuan Yang