With the development of industry and the acceleration of urbanization, we must pay attention to environmental monitoring, because air quality has a paramount impact on human health. The collection and processing of environmental data are prerequisites for environmental monitoring and pollution warning. As a novel technology, the Internet of Things (IoT) is considered to be a pollution monitoring solution, which attracts attention from both academia and industry. In addition, cloud computing makes up the lack of computing resources and energy capacity in environmental monitoring terminals devices. In an environmental monitoring system based on IoT and cloud computing technology, the environmental data collected by the sensors are simply processed by the IoT devices and uploaded to the cloud server. Subsequently, the cloud server processes and analyzes the environmental data uploaded from all sensors. However, with the decrease of granularity of environmental monitoring and the increase of data analysis application amount, the centralized cloud computing architecture is facing challenges, such as high latency, low coverage, and lagged data transmission [
1].
Edge computing, as a new computational paradigms, reduces the application completion latency and the energy consumption of data transmission by distributing cloud resources closer to where data generation [
2]. Fang et al. [
3] has proven that edge computing could enhance the real-time performance of service completion, with reducing computing load and power consumption in the cloud, by offloading computing request to edge servers closer to the IoT device for execution. The authors in [
4] proposed a generalized logical sphere(GLS) modeling scheme in order to avoid servers overload.
In edge-computing based environmental monitoring system, sensors collect the environmental data in real time and transmit them to the edge computing server in order to execute necessary processing and analysis, which reduces system energy consumption and network traffic [
5]. The accuracy of environmental monitoring is affected by positioning problems of monitoring sensors. To solve this problem, Fei et al. [
6] investigated advanced parameter prediction skills and proposed a a smart collaborative tracking scheme to improve particle filter approaches. A number of previous works have proven that resource allocation strategies could reduce task completion latency and energy consumption [
7,
8,
9]. Fengxian et al. [
10] design a genetic algorithm and particle swarm optimization-based algorithm to solve resource allocation problem. However, the dependency between subtasks was not considered in this study. Non-dominated sorting genetic algorithm II was adopted in order to realize multi-objective optimization to reduce the execution time and energy consumption of edge computing devices in paper [
11]. Songtao et al. [
12] has proven that resource allocation policy is determined by the computing workload of a task and the maximum completion time of its immediate predecessors. In [
13], a smart collaborative automation (SCA) scheme was proposed in order to improve resource usage. Du et al. [
14] proposed an algorithm, which obtained allocation decision via semidefinite relaxation and randomization, but the communication among subtasks are ignored by this work. The authors in [
15] took dependencies among subtasks into account, and proposed an multistage greedy adjustment (MSGA) algorithm to solve the task allocation problem. They minimized the completion time of application by jointly considering the network flows and tasks. Laizhong et al. [
16] focus on the task offloading problem in order to find out the optimal tradeoff between task completion latency and energy consumption. They proposed a modified fast and elitist nondominated sorting genetic algorithm to solve the offloading problem. However, in this work, the tasks are considered to be undivided, which wastes the parallel computing capability of edge computing servers. Authors in [
17] created un-excuted task queue at the MEC server and proposed an online algorithm to allocate resources. Pereira et al. [
18] propose an allocation and management resources mechanism to reduce the model complexity of resource allocation algorithm. The authors in [
19] take the dependency between task into account, and proposed a deep reinforcement learning approach to make offloading decision, the difference among tasks is ignored by this work. Xu et al. [
20] adopt Non-dominated Sorting Genetic Algorithm II to shorten the resource allocating time of the computing tasks and reduce the energy consumption of the edge computing servers, but this work did not consider the dependencies among tasks.
In most previous studies, the particularity of environmental monitoring applications has not been considered. For example, air pollution tracing analysis can only be performed on certain servers with wind direction information of the region. In addition, as the chromosome size increases, the time complexity of the resource allocation algorithm increases exponentially [
10]. In order to reduce the solution space of genetic algorithm, we cluster subtasks in genetic algorithm by k-means algorithm, which makes the resource allocation algorithm converge faster.
This study targets the minimum time cost of task completion in environmental monitoring system by fully utilizing the parallel computing capacity of edge computing. The main contributions of this work are outlined, as follows:
The remainder of this paper is organized, as follows.
Section 2 introduces the system model and formulates the resource allocation problem.
Section 3 introduces the proposed resource allocation algorithm and task scheduling strategy.
Section 4 describes a simulation analysis of the proposed algorithm and compares it with the traditional algorithm.
Section 5 concludes the study.