Research on Key Technologies of Edge Computing

Jiarong Chen

College of Arts and Sciences, Beijing Normal University, No. 18 Jinfeng Road, Zhuhai, China

Keywords: Edge Computing, Task Scheduling and Resource Management, Information System Security, Cloud

Computing, Internet of Things.

Abstract: Edge computing, as a key technology for IoT applications, has become an important driver of digital

transformation due to its ability to provide low-latency and high-bandwidth data processing at the edge of the

network. With the explosive growth of IoT devices, the traditional cloud computing model is facing many

challenges, especially in low-latency, high-bandwidth, real-time data processing scenarios. Edge computing

can significantly reduce data transmission time, reduce network congestion, and improve quality of service

by pushing computing resources and data processing capabilities to the edge of the network. This paper

reviews the key technologies of edge computing, focusing on the research progress in task scheduling and

resource management, as well as information system security. By analyzing the current mainstream methods

and models, this paper summarizes the advantages and limitations of edge computing in task scheduling,

resource management, and security evaluation, etc. Finally, it looks forward to the future research direction

and proposes potential ways to optimize the performance of the edge computing system, including multi-

source data fusion and intelligent scheduling mechanism.

1 INTRODUCTION

The rise of edge computing is driven by the explosive

growth of IoT devices and the resulting demand for

data processing. Traditional cloud computing models

face many challenges when dealing with low latency,

high bandwidth, and massive data processing

scenarios in real time (Alrowaily & Lu, 2018). For

example, vehicle-to-vehicle (V2V) or vehicle-to-

infrastructure (V2I) communication in intelligent

transportation systems requires low latency at the

millisecond level (Wang, 2024) to ensure collision

warnings and traffic flow optimization, while real-

time control and monitoring aspects in industrial

automation require extremely high time-to-moment

data processing, where a slight delay can trigger

production failures (Garg, Singh, Kaur, et al., 2019).

Based on these demands, edge computing has

emerged as a key enabler to drive digital

transformation by pushing computing resources and

data processing capabilities to the edge of the network

to significantly reduce the time and distance of data

transmission, alleviate network congestion, reduce

https://orcid.org/0009-0002-1342-1173

latency, and improve service quality and user

experience (Wu, 2024).

In edge computing networks, the optimization of

task scheduling and resource management directly

affects the overall performance of the system and the

efficiency of resource utilization. Since edge nodes

have limited computational resources, the key to

ensuring efficient processing is how to rationally

distribute tasks based on the characteristics of the

tasks, such as computational requirements, delay

constraints, and priorities. In terms of resource

management, the heterogeneous resources involved

in edge computing (e.g., compute, storage, and

network bandwidth) need to be dynamically allocated

and co-optimized to avoid idle or overloaded

resources and ensure that the tasks can be

successfully executed under resource constraints

(Feng, 2023). This is critical to improving system

stability and reliability, e.g., in smart home scenarios,

where multiple tasks generated by smart devices

require precise scheduling and resource allocation to

ensure the stability of the home network and the

smoothness of user interactions (Mijuskovic,

Chiumento, Bemthuis, et al., 2021).

Chen, J.

Research on Key Technologies of Edge Computing.

DOI: 10.5220/0013677500004670

Paper published under CC license (CC BY-NC-ND 4.0)

In Proceedings of the 2nd International Conference on Data Science and Engineering (ICDSE 2025), pages 26-32

ISBN: 978-989-758-765-8

At the same time, the security of edge computing

information systems has become a challenge that

cannot be ignored. Due to its complex architecture,

limited resources, and dynamic and changing

application scenarios, the security posture of edge

computing is severe and complex. Therefore, it is of

great importance to construct an accurate security

evaluation model. Traditional security evaluation

methods often have limitations in edge computing

scenarios, for example, the single-assignment method

is easily disturbed by subjective factors, the first-

order gray clustering method has fuzzy degree

determination, and the fuzzy evaluation model is

computationally expensive and poorly adapted (Guo,

2024). Therefore, designing a security evaluation

model with high accuracy, high efficiency and

dynamic adaptability can more accurately assess the

security status of the system, timely warn potential

risks, reasonably allocate security resources and

optimize protection strategies. This will provide solid

theoretical and technical support for the application

of edge computing in key areas such as finance,

healthcare, energy, etc., ensure the security of data

assets and privacy, and promote the in-depth

application of edge computing technology in various

industries (Chen, 2020).

The application of edge computing in IoT

significantly improves the speed and efficiency of

data processing, but it also brings new challenges in

task scheduling, resource management, and

information security. The next section addresses these

key areas. Chapter 2 details task scheduling and

resource management strategies for edge computing,

including prediction-based resource allocation,

priority-aware task scheduling, and bursty traffic

response schemes, and focuses on information system

security, introducing several efficient security

evaluation models and their application scenarios.

Chapter 3 discusses the commonly used evaluation

criteria and datasets in experiments. Chapter 4

summarizes the current research results, and the last

two chapters point out future directions and potential

research avenues. Through these chapters, this thesis

aims to lead the reader to a comprehensive

understanding of the current state of edge computing

technology and its future development.

2 OVERVIEW OF MAINSTREAM

APPROACHES IN RECENT

YEARS

2.1 Task Scheduling and Resource

Management Aspects of Edge

Computing

·Edge-cloud collaborative task offloading scheme

based on resource utilization prediction: machine

learning models (e.g., LSTM, GRU) are used to

predict the future resource utilization and combined

with Deep Deterministic Policy Gradient Algorithm

(DDPG) to decide the task offloading strategy. When

the edge nodes are resource-constrained, the tasks are

intelligently offloaded to the cloud or other edge

nodes to balance the load, improve resource

utilization, and reduce task latency.

· Multilateral cloud collaborative task scheduling

strategy based on task priority awareness: through the

multi-intelligent body deep Q network (MADQN)

framework, each intelligent body is responsible for

the edge node scheduling, collaborating through

information sharing and learning to optimize

scheduling strategy according to the reward

mechanism to ensure that high-priority tasks are

prioritized and resources are reasonably allocated to

reduce latency and improve the task completion rate,

which is suitable for multi-user multi-task resource

competition scenarios.

·Joint arithmetic slicing strategy based on business

intent constraints: maps business intent to network

state demand, generates arithmetic slicing strategy by

means of Proximal Policy Optimization (PPO)

algorithm, allocates resources to different business

demands, realizes dynamic and flexible resource

allocation, reduces high load blocking rate and intent

violation rate, and improves resource utilization

efficiency and business adaptability.

·Network survivability assurance solution based on

burst traffic prediction: Uses a machine learning

model to predict burst traffic, sets protection

thresholds, and schedules traffic using a particle

swarm optimization algorithm based on policy

gradient (PSO-PG). As network traffic fluctuates, it

plans ahead and precisely allocates resources

according to the predicted feature variables to ensure

low-latency processing of tasks, control the blocking

rate, maintain network stability and survivability, and

withstand unexpected traffic surges (Sun, 2024).

Research on Key Technologies of Edge Computing

2.2 Information System Security

Evaluation Model Aspects of Edge

Computing

·High-precision security evaluation grading model

based on combinatorial assignment gray clustering

(C-SG model): through security policy

decomposition and security index mapping, construct

a pool of security evaluation indexes, propose

subject-objective combinatorial assignment method,

improve the first-order gray clustering method, and

construct a second-order gray clustering security

evaluation method, so as to improve the accuracy of

security evaluation grading.

· Efficient safety evaluation model based on

Spearman's PCA distinction combination screening

(PSDC-CVF model): improve the PCA indicator

screening method, propose the PCA-S indicator

screening method, and combine the indicator

distinction screening method to reduce resource

consumption, improve the indicator screening

efficiency, and achieve a better balance between

efficiency and accuracy.

·Dynamic safety evaluation model based on time-

varying motorized screening of indicator validity

within gray classes (GFCIV-CGTOPSIS model):

separating subjective scoring type indicators and

objective collection type indicators, improving gray

rough set screening method, forming gray F-

statistical clustering screening method of indicator

validity within gray F-statistical clustering, ensuring

high efficiency of the dynamic evaluation model, and

using gray TOPSIS method to dynamically evaluate

the safety state of the system (Guo, 2024). Evaluation

of system security status using gray TOPSIS method

(Guo, 2024).

3 FREQUENTLY USED DATA

SETS AND EVALUATION

CRITERIA

3.1 Edge Computing Task Scheduling

and Resource Management

3.1.1 Datasets

Mostly simulation data sets are used to construct an

edge computing network simulation environment

with a three-layer architecture (terminal layer, edge

layer, cloud layer). The terminal layer covers all kinds

of mobile and sensor devices, randomly generating

data volume and task request time according to the

task characteristics of the device; the edge layer

consists of performance heterogeneous servers; the

cloud layer is a powerful and stable computing node.

The task request includes key characteristics such as

computation demand, latency constraints, task size,

etc. Meanwhile, historical time-series data such as

edge node CPU utilization, task queue length,

network communication latency, bandwidth

utilization, etc. are recorded for model training and

testing, which comprehensively simulates the

dynamic change of task load and complex scenarios

of network state (Yu, 2024).

3.1.2 Evaluation Criteria

·Average task latency: reflects the task processing

speed, which directly affects user experience and

system real-time responsiveness.

·Offload failure rate: reflects the reliability of the

task offloading strategy, which must be strictly

controlled at a low level to ensure smooth task

execution.

·Resource utilization rate: evaluates the degree of

resource utilization of edge nodes, and efficient

utilization can avoid wasting idle resources.

· Task completion rate: emphasizes the

comprehensive ability of the system to handle tasks,

and a high completion rate ensures that task

requirements are effectively met.

3.2 Security Evaluation of Edge

Computing Information Systems

3.2.1 Datasets

The data set is obtained from the fusion of actual

system monitoring data and simulated security event

scenario data, collecting data on system architecture,

device configuration, network connection, user

access behavior, application operation logs, etc.;

simulating various types of network attacks, data

leakage risk scenarios, and system failure status data,

comprehensively covering edge information system

security elements and potential risk dimensions.

3.2.2 Evaluation Criteria

·Security evaluation accuracy: accurately measure

the accuracy of system security status and risk level

determination, and quantify the accuracy of the

evaluation model in identifying and classifying

security events by using confusion matrix, accuracy

rate, recall rate, F1 value and other indicators.

ICDSE 2025 - The International Conference on Data Science and Engineering

· Model efficiency: Focusing on resource-

constrained scenarios, researchers evaluate the degree

of computational resource consumption by model

computation time and memory usage, and shorten the

evaluation cycle and reduce the dependence on

hardware resources with efficient models.

·Dynamic evaluation efficiency: Given the spatial

and temporal dynamic changes of the system, the

model is tested for its ability to dynamically track and

predict the security status, such as calculating the

deviation of dynamic security indicators and

predicting the accuracy of events, to ensure that the

model dynamically adapts to the evolution of the

system and warns of the risk trend in real time (Deng,

Liu, Wang, et al., 2020).

4 ANALYSIS OF CURRENT

RESULTS

4.1 Results of Edge Computing Task

Scheduling and Resource

Management Strategy

· Edge-cloud cooperative task offloading scheme

based on resource occupancy prediction: in a

simulation environment with 20 edge nodes and 2

cloud servers covering 10 square kilometers of

network topology, the performance of processing

500-2000 task requests per round is excellent. The

average task latency is significantly reduced

compared with edge-only or cloud-only processing,

the offload failure rate is controlled within 5.1%, and

the resource utilization rate of edge nodes is increased

to about 70%, effectively relieving the resource

bottleneck, proving its significant advantages in

optimizing task latency, ensuring offload reliability,

and improving resource efficiency.

·Multi-cloud cooperative scheduling strategy based

on task priority awareness: Tested in static and

dynamic load scenarios with remarkable results. The

average processing latency of tasks under dynamic

load is stabilized at 28.4 milliseconds, the completion

rate of tasks exceeds 98%, and the resource utilization

rate of severe resource-guzzling scenarios still

exceeds 60%. This strategy accurately detects task

priority, efficiently schedules resources, and

significantly improves system service quality and

resource utilization performance.

· Common arithmetic slicing strategy based on

business intent constraints: Excellent results in

single-service (e.g., live video streaming) and mixed-

services (e.g., IoT and VR in parallel) scenarios. Task

blocking rate is as low as 0.08% under heavy load,

and intent violation rate is within 8.43%. Dynamic

arithmetic slicing improves resource utilization,

reduces network congestion, effectively improves

service processing efficiency, and guarantees service

quality.

·Network survivability guarantee solution based on

burst traffic prediction: Under the simulation of low-

burst, high-burst and multi-node failure scenarios, the

traffic prediction accuracy reaches 94.2%, the

average task completion delay is only 137

milliseconds, and the blocking rate of high-burst

traffic is continuously controlled within 9.2%. The

solution significantly enhances the network's ability

to withstand sudden impacts and guarantees the

timeliness of task processing and network stability

(Sun, 2024).

4.2 Edge Computing Information

System Security Evaluation Model

Results

·A high-accuracy security evaluation grading model

based on combined-assignment gray clustering (C-

SG model): a multi-level system with 8 first-level and

42 second-level indicators is constructed to improve

the global and scalability. Experiments show that the

difference coefficient of assignment degree is

increased by more than 40%, and the first and last

consistency rate is more than 85%, which

significantly improves the accuracy of safety

evaluation and grading, and accurately determines the

safety level of the system.

· Fuzzy comprehensive and efficient safety

evaluation model based on indicator selection

(PSDC-CVF model): 42 core safety indicators are

optimized and selected to build the evaluation system,

resource consumption is reduced by 67%, and the

offset of fuzzy evaluation results is only 1.8%. While

ensuring the accuracy of fuzzy evaluation, it also

greatly improves the efficiency and efficiently

handles the security evaluation of edge information

systems.

· Dynamic security evaluation model based on

indicator motorized screening (GFCIV-CGTOPSIS

model): The test of telematics edge system shows that

compared with the traditional gray rough set

screening method, the reasonableness of indicator

weight distribution is improved by more than 30%,

the deviation of evaluation results is reduced by about

40%, and the time consumption of the same resource

is reduced by 25%. Dynamic adaptation system

Research on Key Technologies of Edge Computing

changes accurate screening indicator real-time

evaluation, greatly improve the dynamic safety

evaluation accuracy and timeliness (Guo, Lu, Tian, et

al, 2023).

5 CHALLENGES AND OUTLOOK

5.1 Edge Computing Task Scheduling

and Resource Management

Edge computing task scheduling and resource

management strategies face several key issues, the

first of which is the limitation of resource prediction

accuracy. Due to the interference of complex factors

in reality, it is often difficult to accurately predict

resource utilization by relying only on historical data

and traditional machine learning models, which

affects the scientific nature of task offload scheduling

decisions and system performance optimization. To

solve this problem, feature dimensions can be

expanded by fusing data from multiple sources (e.g.,

weather, holidays, geographic location, etc.), and

advanced architectures (e.g., Transformer, BERT,

etc.) can be used to capture spatiotemporal

dependencies and build hybrid models to improve the

accuracy of resource prediction in complex scenarios

(Cen, Hu, Cai, et al., 2022).

Second, edge computing faces a dilemma in large-

scale task scheduling, especially when using the

MADQN framework, the computational cost of

handling massive tasks skyrockets, the efficiency

drops sharply, and it is difficult to meet real-time

requirements, which in turn limits the system's task

processing scale and efficiency improvement. In this

regard, a hierarchical partition scheduling

architecture can be introduced to stream processing

based on task characteristics and resource hierarchy,

and at the same time combined with distributed

training to accelerate the learning convergence of

MADQN intelligences, thus improving the efficiency

of large-scale task processing and system scalability.

In addition, there is a lag problem in the dynamic

adaptation of arithmetic slices, and the PPO algorithm

is slow to adapt to sudden changes in business

demand, which affects the delay of business

processing and fluctuations in service quality. To

solve this problem, it is proposed to construct an

adaptive arithmetic slicing mechanism, design an

intelligent module that monitors changes in business

demand in real time, and adaptively adjust the

arithmetic allocation strategy combined with

reinforcement learning to dynamically optimize the

slicing to ensure smooth business and improve

resource efficiency.

Predicting burst traffic is also a current problem.

Existing models are difficult to accurately capture

extreme burst traffic, and relying on predictive

models when burst traffic occurs leads to scheduling

delays that affect task timeliness and exacerbate

network congestion. Therefore, it is critical to

strengthen the synergy between traffic prediction and

contingency scheduling. Real-time monitoring and

predictive models can be integrated to activate an

early warning mechanism when unexpected traffic

thresholds are encountered, combined with

reinforcement learning for online optimal scheduling

to ensure uninterrupted task processing and maintain

smooth network performance.

Finally, current models are less generalizable and

the evaluation system is too one-sided, with many

models based on specific simulations and lacking

validation on multi-industry and multi-regional data,

resulting in weak generalizability. The evaluation

system also often ignores energy consumption and

costs, which affects the long-term sustainable

operation and market competitiveness of the system.

Therefore, it is necessary to expand the validation

scope of the model, collect multi-industry and multi-

region data for training and optimization, and

improve its generalizability. At the same time, a

multidimensional evaluation system should be

established to introduce energy consumption (e.g.,

computing and communication energy consumption)

and cost (e.g., equipment and operation cost)

indicators, and the dynamic environment should be

simulated to evaluate the performance of the system

in the whole life cycle, so as to provide

comprehensive and accurate decision support for the

optimization and operation of the system (Luo, Hu,

Li, et al., 2021).

5.2 Security Evaluation Model of Edge

Computing Information System

In the security evaluation model of edge computing

information system, one of the first problems is the

difficulty of dynamic development of the index

system. With the rapid development of edge

computing technology and the continuous

improvement of security standards, the existing

security evaluation index system often fails to reflect

the new risks and protection needs in a timely

manner. For example, with the emergence of new IoT

protocols, the lack of security indicators leads to a lag

in the timeliness of the model, which affects the

accurate metrics of the system's security status, thus

ICDSE 2025 - The International Conference on Data Science and Engineering

weakening the effectiveness of security decision-

making. To this end, a dynamic update mechanism

for adaptive indicators can be constructed to track the

latest developments in technical standards and

security research in real time, and iteratively update

the indicators according to emerging risks and

technologies to ensure that the model can adapt to the

latest security needs. In addition, machine learning

methods are used to select key indicators that have a

significant impact on system security, and to ensure

that the indicator system has good scalability and can

be continuously enriched and improved as technology

evolves.

Another problem is the data collection dilemma

and quality bottleneck. The complex architecture and

limited resources of edge computing systems make it

difficult to collect safety data, and expert evaluation

is highly subjective, and there are omissions and

errors in objective collection, resulting in uneven data

quality and limited scale, which affects the accuracy

and generalization of the evaluation model. For

example, in industrial control networks, it is

impossible to comprehensively collect key security

parameters, which leads to the model misjudging

risks and affecting the allocation of security

resources. To solve this problem, a multifaceted

fusion data collection strategy can be innovated to

combine active detection and passive listening means

to comprehensively collect safety data. In addition,

blockchain technology is used to ensure the

trustworthiness and non-tamperability of the data,

improve the security and reliability of the data, and

expand the scope of data collection by introducing a

third-party data source to improve the quality of the

data and the generalization ability of the model.

Finally, the security evaluation model of edge

computing information system also faces the problem

of complexity of effectiveness validation. Existing

models lack unified validation standards and have a

single metric, which makes it difficult to

comprehensively measure the performance

advantages and disadvantages of the model, and

hinders the iterative improvement and application

promotion of the technology. To this end, a

comprehensive and standardized validation and

evaluation system can be established, and common

validation guidelines can be formulated to ensure that

different models can be effectively compared and

integrated with each other. At the same time, a

multidimensional evaluation index system can be

established, covering multiple dimensions such as

performance, safety, economic cost, etc., to

comprehensively evaluate the advantages and

disadvantages of the models. On this basis, a public

testing platform is established to provide a

standardized testing environment and data set, which

facilitates the validation and optimization of the

models by researchers and enterprises, and promotes

the rapid development and application promotion of

the technology (Ma & Li, 2018).

6 CONCLUSIONS

This paper provides a comprehensive overview of the

key areas of edge computing technology, especially

the latest research progress in task scheduling and

resource management, and information system

security. First, in terms of task scheduling and

resource management, this paper analyzes the

challenges facing edge computing, such as resource

occupancy prediction, task priority awareness, and

arithmetic slicing, and proposes solutions based on

machine learning, deep learning, and multi-

intelligence body collaboration. These solutions

effectively improve resource utilization, reduce task

latency, and enhance system reliability. Second, in

the field of information system security, with the

rapid development of edge computing, security issues

are becoming more and more prominent. Aiming at

the dynamic security risks in edge computing

environment, this paper explores new security

evaluation models such as combined-empowerment

gray clustering, PCA indicator screening, etc., which

have achieved remarkable results in improving the

accuracy and efficiency of security evaluation.

However, there are still some challenges in the

existing research, such as insufficient resource

prediction accuracy and difficulty in dynamically

evolving system security evaluation metrics. In the

future, edge computing will face problems such as

large-scale task scheduling, traffic burst prediction,

and system generalization capability, which require

the fusion of multi-source data, the improvement of

model adaptability, and the exploration of efficient

security policies. In conclusion, the research of edge

computing technology is still in continuous

development, and in the future, it will better serve the

Internet of Things, Industrial Internet and other fields

through intelligent and dynamic optimization

solutions to promote the process of digital

transformation.

Research on Key Technologies of Edge Computing

REFERENCES

Alrowaily, M., & Lu, Z. 2018. Secure edge computing in

IoT systems: review and case studies. 2018 IEEE/ACM

symposium on edge computing (pp. 440-444). IEEE.

Cen, B., Hu, C., Cai, Z., Wu, Z., Zhang, Y., Liu, J., & Su,

Z. 2022. A configuration method of computing

resources for microservice-based edge computing

apparatus in smart distribution transformer area.

International Journal of Electrical Power & Energy

Systems, 138, 107935.

Chen, X. 2020. A security integration model for private data

of intelligent mobile communication based on edge

computing. Computer Communications, 162, 204-211.

Deng, X., Liu, J., Wang, L., & Zhao, Z. 2020. A trust

evaluation system based on reputation data in mobile

edge computing network. Peer-to-Peer Networking and

Applications, 13, 1744-1755.

Feng, C. 2023. Task Scheduling and Resource Management

in Edge Computing Networks (Ph.D. dissertation, Xi'an

University of Electronic Science and Technology).

Ph.D.

Garg, S., Singh, A., Kaur, K., Aujla, G. S., Batra, S.,

Kumar, N., & Obaidat, M. S. 2019. Edge computing-

based security framework for big data analytics in

VANETs. ieee network, 33(2), 72-81.

Guo, Z. 2024. Research on Information System Security

Evaluation Model for Edge Computing (Ph.D.

dissertation, Beijing University of Posts and

Telecommunications). Ph.D.

Guo, Z., Lu, Y., Tian, H., Zuo, J., & Lu, H. 2023. A security

evaluation model for multi-source heterogeneous

systems based on IOT and edge computing. Cluster

Computing, 1-15.

Luo, Q., Hu, S., Li, C., Li, G., & Shi, W. 2021. Resource

scheduling in edge computing: a survey. IEEE

Communications Surveys & Tutorials, 23(4), 2131-

2165.

Ma, X., & Li, X. 2018. Trust evaluation model in edge

computing based on integrated trust. 2018 International

Conference on Algorithms, Computing and Artificial

Intelligence (pp. 1-6).

Mijuskovic, A., Chiumento, A., Bemthuis, R., Aldea, A., &

Havinga, P. 2021. Resource management techniques

for cloud/fog and edge computing: an evaluation

framework and classification. sensors, 21(5), 1832.

Sun, Z. 2024. Research on Task Scheduling and Resource

Management Strategies in Edge Computing Networks

(Ph.D. dissertation, Beijing University of Posts and

Telecommunications). Ph.D.

Wang, G. 2024. Research on key technology of multi-

access edge computing based computing service for

Telematics (Ph.D. dissertation, Beijing University of

Posts and Telecommunications, Beijing, China). Ph.D.

Wu, Z. 2024. Research on key technology of blockchain-

based offloading for edge distributed computing (Ph.D.

dissertation, Beijing University of Posts and

Telecommunications). Ph.D.

Yu, X. 2024. Research on task scheduling and resource

management for heterogeneous edge computing with

cloud-side-end collaboration (Ph.D. dissertation,

Beijing University of Posts and Telecommunications).

Ph.D.

ICDSE 2025 - The International Conference on Data Science and Engineering