Computation-intensive Tasks Research Articles

Task Offloading Optimization in NOMA-Enabled Dual-Hop Mobile Edge Computing System Using Conflict Graph

Resource allocation is investigated for offloading computational-intensive tasks in dual-hop mobile edge computing (MEC) system. The envisioned system has both the cooperative access points (APs) with the computing capability and the MEC servers. A user-device (UD), therefore, first uploads a computing task to the nearest AP, and the AP can either locally process the received task or offload to MEC server. To utilize the radio resource blocks (RRBs) in the APs efficiently, we exploit the non-orthogonal multiple access (NOMA) for offloading the tasks from the UDs to the AP(s). In order to investigate the trade-off between latency and energy consumption, this work considers minimizing a weighted-sum that consists of latency and energy consumption, subject to UDs’ rate threshold, tasks’ time-delay, computational frequency scaling, and transmit power allocation constraints. With a joint consideration of all such factors, the problem is NP-hard and its global optimal solution is computationally intractable. A graph-theoretical approach is employed to solve the problem efficiently. Specifically, a novel joint MEC graph-based approach is devised, which solves the scheduling among the UDs, APs, and RRBs, the transmit power control, and the local computational frequency scaling problem(s) jointly. The joint MEC approach achieves near-optimal performance with high computational complexity. To strike a suitable balance between the performance and computational complexity of the resource allocation, a low complexity, yet efficient, pruning graph approach is also devised. The efficiency of the proposed graph-based approaches over several benchmark schemes is verified via extensive simulations.

A Review of the Current Task Offloading Algorithms, Strategies and Approach in Edge Computing Systems

Task offloading is an important concept for edge computing and the Internet of Things (IoT) because computationintensive tasks must be offloaded to more resource-powerful remote devices. Task offloading has several advantages, including increased battery life, lower latency, and better application performance. A task offloading method determines whether sections of the full application should be run locally or offloaded for execution remotely. The offloading choice problem is influenced by several factors, including application properties, network conditions, hardware features, and mobility, influencing the offloading system’s operational environment. This study provides a thorough examination of current task offloading and resource allocation in edge computing, covering offloading strategies, algorithms, and factors that influence offloading. Full offloading and partial offloading strategies are the two types of offloading strategies. The algorithms for task offloading and resource allocation are then categorized into two parts: machine learning algorithms and non-machine learning algorithms. We examine and elaborate on algorithms like Supervised Learning, Unsupervised Learning, and Reinforcement Learning (RL) under machine learning. Under the non-machine learning algorithm, we elaborate on algorithms like non(convex) optimization, Lyapunov optimization, Game theory, Heuristic Algorithm, Dynamic Voltage Scaling, Gibbs Sampling, and Generalized Benders Decomposition (GBD). Finally, we highlight and discuss some research challenges and issues in edge computing.

Massive MIMO-NOMA Based MEC in Task Offloading for Delay Minimization

Mobile edge computing (MEC) has been considered a promising technology to reduce task offloading and computing delay by enabling mobile devices to offload their computation-intensive tasks. Non-orthogonal multiple access (NOMA) is regarded as a promising method of increasing spectrum efficiency, while Massive multiple-input multiple-output (MIMO) can support a larger number of users for simultaneous offloading. These two technologies can effectively facilitate offloading and further improve the performance of MEC systems. In this work, we propose a NOMA and Massive MIMO assisted MEC system for delay-sensitive applications. Our objective is to minimize the overall computing and transmission delay under users’ transmit power and MEC computing capability. Through the pairing scheme for Massive MIMO-NOMA, the users with the higher channel gain can offload all their data, while the users with the lower channel gain can offload a portion of their data to the MEC. Performance results are provided regarding to the sum data rate and overall system delay compared with the orthogonal multiple access (OMA)-MIMO based and Massive MIMO (M-MIMO) based MEC systems.

An Energy-Efficient Domain-Specific Architecture for Regular Expressions

Regular Expressions (REs) are a computational kernel widely used for finding patterns in data in compute-intensive tasks such as genomic markers research, signature-based detection, and database query. Although flexible on the set of searched REs, software-based solutions cannot fulfill latency or throughput requirements to analyze massive data volumes at a given power budget. For this reason, many approaches exploit hardware accelerators as an offloading engine for REs matching. Indeed, various solutions rely on FPGA reconfigurability to embed automata into the reconfigurable fabric. However, this approach leads to time-consuming updates of the REs to search. This work exploits REs as sequences of basic instructions and builds a Domain-Specific Architecture (DSA), called \tirex, for RE matching on FPGAs. Our approach enables the user to change the desired RE at run-time, providing software programmability, flexibility, and specialized hardware mechanisms. Our DSA delivers performance in line with other state-of-the-art hardware approaches, while providing remarkable flexibility and we underline the importance of energy efficiency for these computations. We compared with multiple state-of-the-art software obtaining remarkable performance while achieving noticeable results with a better energy efficiency that ranges from 3x to 490x with our multi-core.

A Deep Reinforcement Learning-Based Offloading Scheme for Multi-Access Edge Computing-Supported eXtended Reality Systems

In recent years, eXtended Reality (XR) applications have been widely employed in various scenarios, e.g., health care, education, manufacturing, etc. Such application are now easily accessible via mobile phones, tablets, or wearable devices. However, such devices normally suffer from constraints in terms of battery capacity and processing power, limiting the range of applications supported or lowering Quality of Experience. One effective way to address these issues is to offload the computation tasks to the edge servers that are deployed at the network edges, e.g., base stations or WiFi access point, etc. This communication fashion, also named as Multi-access Edge Computing (MEC), is proposed to overcome the limitations in terms of long latency due to long propagation distance of traditional cloud computing approach. XR devices, that are limited in computation resources and energy, can then benefit from offloading the computation intensive tasks to MEC servers. However, as XR applications are comprised of multiple tasks with variety of requirements in terms of latency and energy consumption, it is important to make decision whether one task should be offloaded to MEC server or not. This paper proposes a Deep Reinforcement Learning-based offloading scheme for XR devices (DRLXR). The proposed scheme is used to train and derive the close-to optimal offloading decision whereas optimizing a utility function optimization equation that considers both energy consumption and execution delay at XR devices. The simulation results show how our proposed scheme outperforms the other counterparts in terms of total execution latency and energy consumption.

Reliable Distributed Computing for Metaverse: A Hierarchical Game-Theoretic Approach

The metaverse is regarded as a new wave of technological transformation that provides a virtual space for people to interact through digital avatars. To achieve immersive user experiences in the metaverse, real-time rendering is the key technology. However, computing intensive tasks of real-time rendering from metaverse service providers cannot be processed efficiently on a single resource-limited mobile device. Alternatively, such mobile devices can offload the metaverse rendering tasks to other mobile devices by adopting the collaborative computing paradigm based on Coded Distributed Computing (CDC). Therefore, this paper introduces a hierarchical game-theoretic CDC framework for the metaverse services, especially for vehicular metaverse. In the framework, idle resources from vehicles, acting as CDC workers, are aggregated to handle intensive computation tasks in the vehicular metaverse. Specifically, in the upper layer, a miner coalition formation game is formulated based on a reputation metric to select reliable workers. To guarantee the reliable management of reputation values, the reputation values calculated based on the subjective logical model are maintained in a blockchain database. In the lower layer, a Stackelberg game based incentive mechanism is considered to attract reliable workers selected in the upper layer to participate in rendering tasks. The simulation results illustrate that the proposed framework is resistant to malicious workers. Compared with the baseline schemes, the proposed scheme can improve the utility of metaverse service provider and average profit of CDC workers.

Edge-Assisted Real-Time Instance Segmentation for Resource-Limited IoT Devices

Instance segmentation exploits the potential of Internet of Things (IoT) devices to perceive environmental semantic information and achieve complex interactions with the physical world. However, IoT devices usually have limited computing and storage resources and cannot afford the intensive computational costs of instance segmentation networks. This article proposes an edge-assisted instance segmentation method for resource-limited IoT devices; it selectively offloads some computation-intensive tasks from IoT devices to edge servers to accelerate the inference processes of instance segmentation networks. To reduce the communication cost caused by computation offloading, a data compression method is proposed to adaptively adjust the downsampling interval based on an attention mechanism. Considering the susceptibility of the computation offloading scheme to network conditions, an adaptive computation offloading strategy that can jointly optimize the offloading point and the data compression ratio is proposed so that the edge-assisted instance segmentation can meet the preset latency requirements while achieving maximal accuracy under volatile network conditions. Extensive experiments are conducted to verify the feasibility and efficiency of our method. The experimental results show that our method induces less latency than existing instance segmentation methods with a slight drop in accuracy.

Privacy-Preserving and Secure Cloud Computing: A Case of Large-Scale Nonlinear Programming

The volume of data is increasing rapidly currently, which poses a great challenge for resource-constrained clients to process and analyze. A promising approach for solving computation-intensive tasks is to outsource them to the cloud to take advantage of the clouds powerful computing ability. However, it also brings privacy and security issues since the data uploaded to the cloud may contain sensitive and private information which should be protected. In this paper, we address this problem and focus on the privacy-preserving and secure outsourcing of large-scale nonlinear programming problems (NLPs) subject to both linear constraints and nonlinear constraints. Large-scale NLPs play an important role in the eld of data analytics but have not received enough attention. In our outsourcing protocol, we rst apply a secure and efcient transformation scheme at the client side to encrypt its private information. Then, we use the reduced gradient method and generalized gradient method at the server side to solve the transformed large-scale NLPs under linear constraints and nonlinear constraints, respectively. We provide security analysis of the proposed protocol, and evaluate its performance. The experimental results show that we can efciently solve the large-scale NLPs and save huge amount of time for the clients.

Deep Reinforcement Learning for Time-Energy Tradeoff Online Offloading in MEC-Enabled Industrial Internet of Things

Mobile edge computing (MEC) has recently emerged as a promising technology to boost the integration ability of sensing, transmission and computation in industrial Internet of Things (IIoT). This paper investigates an MEC-enabled IIoT system, where multiple industrial devices may offload computation-intensive tasks to an edge server through wireless communication. We focus on the online offloading problem to optimize the tradeoff of the task accomplishing time and energy consumption. Time-varying wireless channels, random targeted task data sizes and dynamically changing residual energy as well as adaptively adjusted tradeoff weights make this problem highly challenging. Conventional optimization methods may lead to inefficient or even infeasible solutions. To efficiently tackle this problem, we leverage the deep reinforcement learning (DRL) technology to propose a time-energy tradeoff online offloading algorithm called TETO. In TETO, the online offloading decision policies are empirically learned via a well-designed DRL framework. TETO algorithm incorporates a stochastic strategy, the crossover and mutation technology and a novel feasible suboptimal offloading method to expand the offloading action search space with the provable feasibility guarantee. Extensive experimental results based on a real-world dataset show that, our TETO algorithm performs better than existing baseline algorithms, and obtains near-optimal performance with low CPU execution latency.

MOBILE EDGE COMPUTING ORIENTED MULTI-AGENT COOPERATIVE ROUTING ALGORITHM: A DRL-BASED APPROACH

In the era of 5G/B5G, computing-intensive, delay-sensitive applications such as virtual reality inevitably bring huge amounts of data to the network. In order to meet the real-time requirements of applications, Mobile Edge Computing (MEC) pushes computing resources and data from the centralized cloud to the edge network, providing users with computing offload technology. However, the mismatch between the great computing requirements of computing-intensive tasks and the limited computing power of a single edge server poses a great challenge to computing offload technology. In this paper, a multi-agent cooperation mechanism for MEC and a routing mechanism based on deep reinforcement learning (DRL) are proposed. First of all, a multi-agent cooperation mechanism is proposed to realize the cooperative processing of computing-intensive and delay-sensitive applications, and the task unloading decision-making problem based on multi-agent cooperation is studied. Secondly, the cooperative processing of tasks by multi-agents involves data transmission. Considering the real-time requirements of tasks, this paper proposes an intelligent routing mechanism based on DRL to plan the optimal routing path. Finally, the simulation implementation and performance evaluation of the multi-agent cooperation mechanism and routing mechanism for MEC are carried out. The experimental results show that the intelligent routing mechanism based on DRL and graph neural network is superior to the comparison mechanism in terms of network average delay, throughput and maximum link bandwidth utilization. At the same time, the superiority of graph neural network in model generalization is verified on a new network topology National Science Foundation (NSF) Net. The results of route optimization are applied to the multi-agent cooperation mechanism, and the experimental results show that the mechanism is superior to the comparison scheme in terms of task success rate and average task response delay. The combination of these two mechanisms well solves the problem that it is difficult to deal with computing-intensive and delay-sensitive applications in mobile edge computing because of its limited resources.

Energy-Efficient Task Offloading of Edge-Aided Maritime UAV Systems

This paper considers the autonomous detecting and tracking task of unmanned aerial vehicles (UAVs) in the maritime environment. Due to the high computational complexity of the image-processing task, the strict latency requirement of the tracking task, and the shortage of batteries and computational capability of UAVs, the UAVs can offload the computing-intensive task to the edge computing server (ECS) to reduce the task latency and energy consumption. Even though, the large latency and energy consumption brought by some high-definition image-processing task still make some visual tracking difficult to perform. We first investigate the impact of video frame resolution on computing task size and detection accuracy, and further on task latency and energy consumption. In addition, we optimize the transmission power, the local CPU frequency, the offloading rate of UAVs, and the bandwidth allocation between UAVs. We model the UAVs tracking system as an energy consumption optimization problem with the constraint of task latency. Although the proposed optimization problem is complex, we transform the optimization problem into several subproblems with decoupling and rigorous mathematical proof. By combining the convex optimization with the genetic algorithm, an offloading algorithm with low complexity and good performance is obtained, which is verified by comparing with some benchmark offloading strategies in the numerical results.

A Distributed Fuzzy Optimal Decision Making Strategy for Task Offloading in Edge Computing Environment

With the technological evolution of mobile devices, 5G and 6G communication and users’ demand for new generation applications viz. face recognition, image processing, augmented reality, etc., has accelerated the new computing paradigm of Mobile Edge Computing (MEC). It operates in close proximity to users by facilitating the execution of computational-intensive tasks from devices through offloading. However, the offloading decision at the device level faces many challenges due to uncertainty in various profiling parameters in modern communication technologies. Further, with the increase in the number of profiling parameters, the fuzzy-based approaches suffer inference searching overheads. In this context, a fuzzy-based approach with an optimal inference strategy is proposed to make suitable offloading decision. The proposed approach utilizes Classification and Regression Tree (CART) mechanism at the inference engine with reduced time complexity of <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">O</i> (| <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">V</i> | <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">log</i> <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sub> | <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">L</i> |)), as compared to <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">O</i> (| <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">L</i> | <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">|V |</sup> ) of state-of-the-art, conventional fuzzy-based offloading approaches, and has been proved to be more efficient. The performance of the proposed approach is evaluated and compared with contemporary offloading algorithms in a python-based fog and edge simulator, YAFS. The simulation results show the reduction in average task processing time, average task completion time, energy consumption, improved server utilization, and tolerance to latency and delay sensitivity for the offloaded tasks in terms of reduced task failure rates.

Deep Reinforcement Learning-Based Task Offloading for Parked Vehicle Cooperation in Vehicular Edge Computing

Vehicular edge computing (VEC) has greatly enhanced the quality of vehicle service with low latency and high reliability. However, in some areas not covered by roadside infrastructures or in cases when the infrastructures are damaged or fail, the offloaded tasks cannot have the chance to be performed. Even in the areas deployed with infrastructures, when a large number of offloaded tasks are generated, the edge servers may not be capable of processing them in time, owing to their computing resources constraint. Based on the above observations, we proposed the idea of parked vehicle cooperation in VEC, which uses roadside parked vehicles with underutilized computational resources to cooperate with each other to perform the compute-intensive tasks. Our approach aims to overcome the challenge brought by infrastructure lacking or failure and make up for the shortage of computing resources in VEC. In our approach, firstly, the roadside parked vehicles are managed as different parking clusters. Then, the optimal amount of resources required for each offloaded task is analyzed. Furthermore, a task offloading algorithm based on deep reinforcement learning (DRL) is proposed to minimize the total cost, which is composed of the task execution delay and the energy consumption overhead of the parked vehicles for executing the task. A large number of simulation results show that, compared with other algorithms, our approach not only has the highest task completion execution successful rate, but also has the lowest task execution cost.

Unmanned Aerial Vehicle Computation Task Scheduling Based on Parking Resources in Post-Disaster Rescue

Natural disasters bring huge loss of life and property to human beings. Unmanned aerial vehicles (UAVs) own the advantages of high mobility, high flexibility, and rapid deployment, and are important equipment during post-disaster rescue. However, UAVs usually have restricted battery and computing power. They are not fit for performing compute-intensive tasks during rescue. Since there are widespread parking resources in a city, multiple parked vehicles working together to compute the applications from UAVs in a post-disaster rescue is investigated to ensure the quality of experience (QoE) of the UAVs. To execute uploaded task effectively, surviving parked vehicles within the monitoring range of an UAV are arranged into a cluster as much as possible. Then, the task execution cost is analyzed. Furthermore, a deep reinforcement learning (DRL)-based offloading policy is constructed, which interacts with the environment in an intelligent way to achieve optimization goals. The simulation experiments show that the proposed offloading scheme has a higher task completion rate and a lower task execution cost than other baselines schemes.

Optimal multi-user offloading with resources allocation in mobile edge cloud computing

Mobile cloud computing (MCC), as a prospective computing paradigm, can significantly enhance computation capability and save energy of smart mobile device (SMD) by offloading computation-intensive tasks from resource-constrained SMDs onto the resource-rich central cloud. Compared to central cloud, edge cloud can provide services to nearby SMDs with lower latency. However, the edge cloud may be mobile and its resources are limited to multiple nearby users. Therefore, how to obtain an optimal offloading policy under the constraints of mobility and limited resource remains a challenging issue. In this paper, we aim to minimize the total execution cost of multiple devices by offloading the computations from SMDs onto edge clouds in edge cloud computing (ECC) system. By considering the mobility of SMDs and edge clouds, we first formulate the total cost minimization problem under the constraints of application completion deadline and connection time between SMDs and edge clouds as well as the limited computation resource of both edge clouds and SMDs. Then, by solving the minimization problem, we propose an optimal offloading selection strategy based on game model, and an edge cloud payoff competition algorithm to optimally allocate edge cloud resource to SMDs to achieve the minimum total execution cost. Experimental results show that our offloading strategy can effectively reduce energy consumption and application completion time compared with the state-of-the-art methods.

An Evaluation of Autoencoder Neural Network Role in Iot Edge Computing

With rapid increase in numbers of connected Internet of Things (IoT) devices, huge amount of data is generated and sent to Cloud Computing nodes to be stored and analysed. Cloud computing is an effective paradigm for storage and data analysis since IoT devices are restricted machines in terms of energy, computation power and storage. Despite the advantages of cloud computing, it causes network congestion and latency due to generally located at long distances. Besides, security and privacy issues are also drawbacks of the cloud. Edge Computing is a promising system to eliminate the flaws of cloud computing by getting computational power closer to data sources. Edge Computing has more computation power than IoTD but lower than cloud computing. Although the deficiencies of cloud computing decrease with edge computing, they are not completely eliminated because computation intensive tasks still should be sent from edge to cloud resources. Since Autoencoder is an unsupervised neural network technique that learns to efficiently encode/compress input data and learns to efficiently decode it as closer to the original input, it is an ideal candidate for reducing data traffic and latency in edge computing and cloud computing. Instead of sending all data to the cloud, the data of bottleneck hidden layers in which input data is encoded are sent from edge to cloud. The compressed data is decoded on the cloud to reconstruct the original input to be analysed and learnt. In this paper, we investigate the studies using AE in edge computing and their performance implications with respect to network traffic and delay. The performance results of the proposals that have used autoencoder between edge and cloud layer are evaluated in terms of eliminating big data, network traffic and accuracy.

Two-tier MPC architecture for AGVs navigation assisted by edge computing in an industrial scenario

The next generation of mobile networks may be a bridge to move industrial control to edge computing. Embedded controllers physically installed in the industrial process can be replaced by software remotely executed as a service. In this context, Automated-Guided Vehicles (AGVs) are industrial agents that can benefit from this scenario by offloading computing-intensive tasks remotely to the edge. Additionally, an AGV control system based on remote software promotes flexibility, an essential requirement for Industry 4.0. Considering the remote control of AGV, the mobile robot may traverse areas where the signal is degraded, increasing the risks of collisions and accidents due to connection failures. Previous works suggest the adoption of Model Predictive Control (MPC) to control mobile robots in the occurrence of delays and packet losses. In this article, we propose a two-tier architecture of MPCs, one executed at the edge to plan the trajectory of multiple AGVs globally and the other executed individually in each AGV to keep it on the planned track. In simulations performed in an edge computing environment using a robot simulator, the vehicles follow the planned trajectory even with network degradation. In our simulated scenarios with delays and packet loss, predicted actions planned by the MPC on the edge avoided collisions between the vehicles.

Solving Sparse Assignment Problems on FPGAs

The assignment problem is a fundamental optimization problem and a crucial part of many systems. For example, in multiple object tracking, the assignment problem is used to associate object detections with hypothetical target tracks and solving the assignment problem is one of the most compute-intensive tasks. To enable low-latency real-time implementations, efficient solutions to the assignment problem is required. In this work, we present Sparse and Speculative (SaS) Auction, a novel implementation of the popular Auction algorithm for FPGAs. Two novel optimizations are proposed. First, the pipeline width and depth are reduced by exploiting sparsity in the input problems. Second, dependency speculation is employed to enable a fully pipelined design and increase the throughput. Speedups as high as 50 × are achieved relative to the state-of-the-art implementation for some input distributions. We evaluate the implementation both on randomly generated datasets and realistic datasets from multiple object tracking.

DeTTO: Dependency-Aware Trustworthy Task Offloading in Vehicular IoT

In this paper, we investigate the dependency-aware trustworthy task offloading problem (DeTTO), especially in an IoT-enabled vehicular network, where a large computation-intensive task offloaded from a vehicle is fragmented into multiple subtasks and then offloaded to multiple trusted nodes. First, we formulate the task offloading problem as a graph optimization problem intending to find an optimal set of trustworthy nodes for offloading the subtasks. We aim to minimize the task completion delay and energy consumption, while satisfying the dependency relations between the subtasks and the trust requirements of the tasks. We consider three types of dependency structures – fully independent task, fully dependent task, and partially dependent task. For a fully independent task with no dependency between the subtasks, we propose a greedy algorithm to get the optimal set of nodes for task offloading. After showing the NP-hardness of solving the dependent task offloading problem, we propose a two-fold efficient heuristic approach for the tasks with all dependent subtasks. We adopt the solution approaches used by the first two types of tasks for a partially dependent task. Through simulation experiments, we analyze the performance of the proposed algorithms for three types of intra-task dependencies. The experimental results show that the proposed algorithms significantly reduce the delay and energy consumption, when compared to the benchmark schemes.

SmartVM: A Smart Contract Virtual Machine for Fast On-Chain DNN Computations

Blockchain-based artificial intelligence (BC-AI) has been applied for protecting deep neural network (DNN) data from being tampered with, which is expected to further boost trusted distributed AI applications in many fields. However, due to smart contract execution environment architectural defects, it is challenging for previous BC-AI systems to support computing-intensive tasks on-chain performing such as DNN convolution operations. They have to offload computations and a large amount of data from blockchain to off-chain platforms to execute smart contracts as native code. This failure to take advantage of data locality has become one of the major critical performance bottlenecks in BC-AI system. To this end, in this article, we propose SmartVM with optimization methods to support on-chain DNN inference for BC-AI system. The key idea is to design and optimize the computing mechanism and storage structure of smart contract execution environment according to the characteristics of DNN such as high computational parallelism and large data volume. We decompose SmartVM into three components: 1) a compact DNN-oriented instruction set to describe computations in a short number of instructions to reduce interpretation time. 2) a memory management mechanism to make SmartVM memory dynamic free/allocated according to the size of DNN feature maps. 3) a block-based weight prefetching and parallel computing method to organize each layer's computing and weights prefetching in a pipelined manner. We perform the typical image classification in a private Ethereum blockchain testbed to evaluate SmartVM performance. Experimental results highlight that SmartVM can support DNN inference on-chain with roughly the same efficiency against the native code execution. Compared with the traditional off-chain computing, SmartVM can speed up the overall execution by <bold xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">70×</b> , <bold xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">16×</b> , <bold xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">11×</b> , and <bold xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">12×</b> over LeNet5, AlexNet, ResNet18, and MobileNet, respectively. The memory footprint can be reduced by <bold xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">84%</b> , <bold xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">90.8%</b> , <bold xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">94.3%</b> , and <bold xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">93.7%</b> over the above four models, while offering the same level model accuracy. This article sheds light on the design space of the smart contract virtual machine for DNN computation and is promising to further boost BC-AI applications.