Field Programmable Gate Array (FPGA) has become an efficient accelerator for convolutional neural network (CNN) inference due to its high performance and flexibility. To further improve the performance of CNN inference on FPGAs, an Intellectual Property core (IP core) called Deep Learning Processor Unit (DPU) is released by Xilinx. Unlike previous FPGA-based hardware designs focusing on specific functions or CNNs, the DPU IP supports ample basic functions of deep learning, and the developers can take advantage of DPUs to accelerate CNN inference conveniently. In DPU-based CNN acceleration platform, an encapsulated scheduler plays a crucial role in task scheduling between heterogeneous ARM and multiple DPUs. However, the current scheduler is unsatisfactory because its low schedule efficiency. This paper thus presents a high performance task assignment framework built upon Xilinx hybrid CPU-FPGA MPSoC devices. We first evaluate the main causes of low schedule efficiency problem. Then, we explore the scheduler rules and improve shedule efficiency through purposeful observations and analysis. Finally, we integrate our optimizations, and propose an efficient task assignment framework to maximize performance on DPU-based CNN acceleration platform. Experimental results on Xilinx Zynq UltraScale+ MPSoC zcu104 show that our efficient task assignment framework significantly boosts schedule efficiency for small-scale CNNs (from 36% to 70%), medium-scale CNNs (from 65% to 95%), and large-scale CNNs (from 77% to 99%) compared with original schedule strategy.

Cloud computing and mobile edge computing techniques supply efficient ways to solve the contradiction between the increasing computing and storage demands of portable terminals and the limited capacity. In this paper, we conduct a three-tier hierarchical service system with multiple UEs, multiple MECs, and a single cloud center. It's worth noting that multiple UEs with personalized options generate a large number of different tasks in real time. To deal with this offloading problem, a response ratio offloading strategy (RROS) centered on user preference and real-time nature is designed to make MECs or CC serve as many UEs as possible. Therefore, a MEC-choosing preference list of each UE is created based on its past experiences at first. Then, each MEC iteratively sorts UEs with its ranking in the UEs' preference list. In order to avoid that the first task arriving at MEC occupies too many resources of MEC and cannot achieve global optimization, we also adopt loop iterative sequencing for multiple tasks arriving within a stipulated time. Lastly, by comparing the optimal response ratio on different MECs and CC, multiple MECs and the CC collaborative offload computing tasks of multiple UEs. Experimental results show that the algorithm significantly outperforms conventional techniques.

The large number of mobile-sensing vehicles traveling in cities offer a novel solution to the collection of vast amounts of multimedia data packets. When a vehicle passes through the data center (DC), the collected multimedia data packets will be transmitted to the DC. Due to the mobile characteristic of vehicular sensor networks, the main challenge lies in how to improve the multimedia data delivery ratio and balance the data packet collections. In this paper, in consideration of delay and delivery factors, a novel routing method is proposed to optimize multimedia data collections in mobile sensing vehicles (DDSVs). This method targets at balancing multimedia data collections, improving the delivery ratio of the multimedia data, and reducing the delay ratio in Internet of Things (IoT) networks. In the DDSV scheme, two rules are designed for improving the collection of multimedia data in the IoT. These rules pertain to: 1) data and 2) vehicular priorities. First, different regions hold different priorities of data packet transmission, which can improve the delivery ratio in the suburban areas and reduce the delay ratio. Meanwhile, this scheme is capable of guaranteeing the balance of multimedia data collection. Second, the vehicular priority is proportional to the probability of a vehicle reaching a DC. Therefore, the data should be forwarded to vehicles with higher priorities, that is, the vehicles which are more likely to pass by the DC. By using these two rules, the DDSV scheme can improve the performances of the multimedia data delivery ratio, compared with the conventional optimal vehicular data forwarding scheme. In the simulation experiments, the DDSV scheme utilizes multidatasets of Beijing city, where the average delay for data collection can be decreased by 17.3% in general, and by 41.8% in the suburban areas; the average data delivery ratio can be improved by 16.9% in comparison to the previous studies.

Due to limited on-board resources and the mobility characteristics of vehicles in a multi-access edge computing (MEC)-based vehicular network, efficient task offloading and resource allocation schemes are essential for achieving low-latency and low-energy consumption applications in the Internet of Vehicles (IoV). The spatial distribution of vehicles, influenced by various factors, leads to significant workload variations across MEC servers. In this paper, we address task offloading and resource allocation as a joint optimization problem and propose a Load-Balancing Deep Deterministic Policy Gradient (LBDDPG) algorithm to achieve optimal results. The joint optimization problem is modeled as a Markov Decision Process (MDP), enabling the LBDDPG algorithm to systematically address the challenges of workload imbalance and resource inefficiency. The algorithm incorporates a load optimization strategy to balance workload distribution across MEC servers, mitigating disparities caused by uneven vehicle distributions. The reward function is designed to account for both energy consumption and delay, ensuring an optimal trade-off between these critical factors. To enhance training efficiency, a noise-based exploration strategy is employed, preventing ineffective exploration during the early stages. Additionally, constraints such as computational capacity and latency thresholds are embedded to ensure the algorithm’s practical applicability. Experimental results demonstrate that the proposed LBDDPG algorithm achieves faster convergence and superior performance in terms of energy consumption and latency compared to other reinforcement learning algorithms.

With data transparency and immutability, the blockchain can provide trustless and decentralized services for Internet of Things (IoT) applications. However, most blockchain-IoT networks, especially those with a private blockchain, are built on top of an infrastructure-based wireless network (i.e., using Wi-Fi access points or cellular base stations). Hence, they are still under the risk of Single-Point-of-Failure (SPoF) on the network layer, hindering the decentralization merit, for example, when the access points or base stations get failures. This paper presents an Optimized Link State Routing (OLSR) protocol-based solution for that issue in a private blockchain-IoT application. By decentralizing the underlying network with OLSR, the private blockchain network can avoid SPoF and automatically recover after a failure. Single blockchain connections can be extended to multiple ad hoc hops. Services over blockchain become flexible to fit various IoT scenarios. We show the effectiveness of our solution by constructing a private Ethereum blockchain network running on IoT devices (i.e., Raspberry Pi model 4) with environmental data sensing (i.e., Particular Matter (PM)). The IoT devices use OLSR to form an ad hoc network. The environment data are collected and propagated in transactions to a pre-loaded smart contract periodically. We then evaluate the IoT blockchain network’s recovery time when facing a link error. The evaluation results show that OLSR can automatically recover after the failure. We also evaluate the transaction-oriented latency and block-oriented latency, which indicates the blocks have a high transmission quality, while transactions are transferred individually.