JOET: Sustainable Vehicle-assisted Edge Computing for IoT devices

Abstract

Network-accessible task offloading is for low latency; offline task offloading is for low cost. Offline devices cannot directly access service nodes due to lack of resources. Accordingly, the latter involves more steps and optimization variables such as: where to offload tasks, how to allocate computation resources, how to adjust offloading ratio and transmit power, and such optimization variables and hybrid combination features are highly coupled with each other. In this paper, we first formulate a Mixed Integer Nonlinear Programming Problem (MINLP) for such task offloading under energy and delay constraints. Furthermore, we decompose it into two subproblems so as to efficiently solve the formulated MINLP, and design a low-cost and low-complexity Joint Optimization for Energy Consumption and Task Processing Delay (JOET) algorithm to optimize selection decisions, resource allocation, offloading ratio and transmit power adjustment. We carry out extensive simulation experiments to validate JOET. Simulation results demonstrate that JOET outperforms many representative existing schemes in quickly converge and effective reduction of energy consumption and delay. Specifically, the average energy consumption and the average delay have been reduced by 15.93% and 13.70%, respectively, and the load balancing efficiency has increased by 10.20%.

Introduction

While the network-accessible task offloading pursues quick response algorithms [1], the offline task offloading is impotent [2]. The network-accessible devices with long-distance communication can directly access the offloading destinations, but offline devices tend to have meager resources (communication and computation), such as Internet of Things (IoT) devices with simple hardware and low cost [3], [4], [5]. On the one hand, IoT devices are progressively deployed where data needs to be sensed and collected [3], [5], [6]. On the other hand, the number of vehicles has grown rapidly which can provide more service [7], [8], [9]. According to the IoT Analytics, the number of IoT devices worldwide is expected to reach 22 billion by 2025 [10]. Leveraging the Mobile Vehicle (MV) as a relay to connect IoT devices to servers is a low-cost and feasible way in which IoT devices can upload their data or tasks to servers via MVs when MVs enter the communication range of IoT devices [3], [4], [11]. Such way has a huge sustainability advantage because IoT devices in smart city can be easily and cost-effectively connected to the Internet via the large scale and widespread vehicles [3].

Previous research works show that task offloading in Internet of Vehicles (IoV) chiefly offloads vehicle tasks to edge servers [12], [13], [14]. Howbeit, IoT devices widely deployed in smart cities have a more urgent need for task offloading [15], [16]. Relatively speaking, the MVs have stronger computing and communication capabilities, so they can offload their tasks to edge servers at any time [13], [17]. Since IoT devices are once deployed, they cannot be moved like MVs. Accordingly, tasks from IoT devices can only be offloaded via relays such as MVs unless there are edge servers available for offloading within communication range of IoT devices [4], [17]. These IoT devices are deployed in various fields to monitor and collect data, bringing broad application prospects [18], [19]. Hence, offloading tasks from massive offline IoT devices is more valuable than the aforementioned task offloading for vehicles.

Howbeit, task offloading in such semi-connected networks in IoV is more challenging due to follow reasons.

First, task offloading for offline IoT devices involves more steps and decisions than network-accessible task offloading. However, researchers pour more attention to network-accessible task offloading [1], [20]. With the advancement of vehicular ad-hoc networks or IoV, a distributed task offloading scheme is proposed in [13] to select nearby vehicles with idle computation resources to process tasks in parallel. This communication method is vehicle-to-vehicle and involves only one decision step. An approach is considered in [21] that vehicular tasks can be offloaded to service nodes by jointly optimizing the tasks and wireless bandwidth ratios to minimize the overall response time, and design a binary search and feasibility check algorithm to solve the optimization problem. Task offloading in [13], [21] involves only one decision step. The research in [22] focus on the offloading decision, collaboration decision, computation resource allocation and communication resource allocation problem. The delay-sensitive tasks of users can be computed locally, offloaded to collaborative devices or MEC servers. The method is actually a cooperative one-step offloading. In addition, a distributed multi-hop task offloading decision model is proposed in [23]. The one-step inter-group communication way is used in [23], where the neighboring vehicles in the $k$-hop wireless communication range are selected as the candidate vehicles. In the above approaches, users are either offloaded to their own group or offloaded to the server. Howbeit, offline IoT devices generally cannot actively access the server or even help other devices to complete their tasks due to lack of resources [24], [25]. Accordingly, the IoT task needs to be uploaded to a relay such as the Unmanned Aerial Vehicle (UAV) [2], [26] or the MV [4], [27]. After the relay receives the task, it needs to make a decision whether to complete the task by itself or offload it to the server. Therefore, IoT devices task offloading involves more decision steps than network-accessible task offloading.

Second, task offloading for offline IoT devices encloses many optimization objectives (task processing time, device energy consumption and server load balancing) and coupling variables. The joint admission control and computation resource allocation is proposed in [28] to balance servers load by keeping each server’s task queue length as equal as possible. The work in [29] focuses on multi-input multi-output MEC systems with energy harvesting and studies the computation offloading, where the design objective is to minimize the time average of a weighted sum of energy consumption and execution delay by adopting the Lyapunov method to stabilize the battery energy queue. The studies in [4], [12] integrate server load balancing into task offloading and aim to minimize task processing delay. The above studies all involve only one optimization objective or two optimization objectives. The UAV-assisted MEC system is studied in [26], where a UAV equipped with an MEC server is deployed to server a number of IoT devices in a finite period, which aims to minimize the total energy consumption including communication-related energy, computation-related energy and UAV’s flight energy. The above studies only consider one or two optimization objectives. A UAV-assisted MEC architecture is also proposed in [16] to provide services to IoT devices, in which the UAV provide both communication and computation services to minimize latency, energy consumption and the operation cost of serving all IoT devices. However, the UAV-assisted approach undoubtedly increases the operating cost for widely deployed offline IoT devices. On the other hand, ubiquitous MVs are suitable for meeting the computing demands of the IoT devices compared to UAVs. To better serve offline IoT devices, not only low latency and low energy consumption should be pursued, but sustainable offloading should also be considered.

The study in this paper aims to reduce task processing delay and energy consumption of IoT devices in load balancing while ensuring task completion. Complex limitations are as follows: First, it is necessary to ensure that tasks can only be offloaded to one place; The allocated resources cannot exceed the maximum task processing delay; The sum of the total resources allocated to multiple tasks by an edge server cannot exceed the total amount of resources of the server; Finally, the ratio of the offloaded computation bits does not exceed 1. Thus, offloading IoT tasks needs to control these complex behaviors, that is, optimizing selection decisions, resource allocation, offloading ratio and transmit power adjustment. There is a highly complex coupling between optimization variables and hybrid combination features, so it is extremely challenging to solve such the task offloading problem. This paper formulates the above task offloading problem as a Mixed Integer Non-Linear Programming (MINLP) problem to better solve this problem. In summary, the main contributions of this paper are summarized as follows:

(1) We first propose the task offloading issue for offline IoT devices compared to network-accessible task offloading. We further propose a low-cost and sustainable offload approach by sprawling previous studies. In this approach, ubiquitous mobile vehicles are used to assist the task offloading of resource-poor IoT devices, which is a pioneer scheme. We formulate the problem of minimizing energy consumption and delay as a system utility maximization problem from the perspective of problem solving, and then transform the problem into a MINLP problem with consideration of load balancing.

(2) We propose a Joint Optimization for Energy consumption and Task processing delay (JOET) algorithm to effectively solve the proposed MINLP to minimize energy consumption and delay. Our proposed JOET first determines selection decisions under given resource allocation, offloading ratio and transmit power, and then optimizes the three variables under obtained selection decisions. The optimization solution with less overhead can be found by repeating the above optimization steps until convergence.

The rest of this paper is organized as follows. In Section 2, we review related works. Section 3 introduces the system model and the problem description, while Section 4 presents the problem decoupling and the solution. We present the performance analysis in Section 5. Finally, Section 6 concludes this paper and introduce future work.