Advanced Framework for Multi-Objective Optimization of Computation Offloading in Heterogeneous MEC Environments

Jyothirmai Munjam¹ , Kesavan Gopal² and M. Sailaja¹
1. Jawaharlal Nehru Technological University Kakinada, Kakinada, Andhra Pradesh, India
2. Lovely Professional University, Phagwara, Punjab, India
Correspondence to: Jyothirmai Munjam, jyothimunjamp@gmail.com

DOI: https://doi.org/10.70389/PJCS.100010

Additional information

• Ethical approval: N/a
• Consent: N/a
• Funding: No industry funding
• Conflicts of interest: N/a
• Author contribution: Jyothirmai Munjam, Kesavan Gopal and M. Sailaja – Conceptualization, Writing – original draft, review and editing
• Guarantor: Jyothirmai Munjam
• Provenance and peer-review:
  Unsolicited and externally peer-reviewed
• Data availability statement: N/a

Keywords: Multi-objective optimization, Computation offloading, Enhanced adaptive cat hunt optimization, Directed acyclic graphs, Heterogeneous mec environments.

Peer Review
Received: 18 December 2024
Accepted: 14 March 2025
Version accepted: 1
Published: 28 August 2025

Plain Language Summary Infographic

Introduction

Data-intensive applications such as augmented reality and IoT services have transformed mobile devices (MDs) into critical computational platforms. However, resource constraints—including limited battery life and processing power—hinder their performance. Mobile Edge Computing (MEC) mitigates these issues by enabling task offloading to nearby edge servers, thereby enhancing computational efficiency and reducing latency. Despite these advancements, MEC faces challenges in:

1. Optimizing energy, delay, and resource utilization simultaneously.
2. Scaling solutions across heterogeneous environments.
3. Adapting real-time decision-making under dynamic workloads.

This research introduces the Enhanced Adaptive Cat Hunt Optimization (EACHO) algorithm to address these challenges. The primary contributions of this study include:

1. Leveraging Directed Acyclic Graphs (DAGs) for effective task dependency modeling.
2. Developing adaptive optimization strategies for dynamic MEC environments.
3. Demonstrating superior performance through comparative analysis with existing methods.

Related Work – Advances in Computation Offloading

Numerous studies have explored optimization methods for computation offloading. Key contributions include the following: Zhang et al.¹ introduced a genetic algorithm for load balancing in MEC, focusing on minimizing service delay and enhancing computational efficiency. Liang et al.² explored federated learning-based task offloading, ensuring data privacy and achieving low latency in MEC scenarios. Chen et al.³ proposed a blockchain-integrated model for secure task offloading, prioritizing data integrity and trust among MEC nodes. Luo et al.⁴ developed a fog computing framework to address real-time task execution challenges in smart city applications. Qin et al.⁵ designed a hierarchical optimization model to improve reliability and reduce latency in heterogeneous MEC environments. Wei et al.⁶ suggested a task replication strategy to minimize the risks of MEC server failures, achieving improved fault tolerance.

Wang et al.⁷ investigated deep transfer learning for task scheduling, enabling efficient adaptation to dynamic MEC workloads. Zhang et al.⁸ introduced a multi-channel communication model to optimize bandwidth utilization and reduce interference in MEC. Wang et al.⁹ developed a reinforcement learning approach for energy-aware task offloading in resource-constrained MEC systems. Lee et al.¹⁰ explored collaborative task offloading between MEC servers and cloud platforms to achieve cost-efficient processing. Zhou et al.¹¹ suggested a bio-inspired algorithm to optimize resource allocation and task prioritization in MEC networks. Liu et al.¹² proposed a probabilistic model for deadline-aware offloading, ensuring task completion within specified time frames.

Zhang et al.¹³ presented a hybrid caching mechanism for frequently offloaded tasks to enhance MEC performance. Chen et al.¹⁴ applied swarm intelligence to optimize multi-user task allocation in dense MEC networks. Liu et al.¹⁵ developed an energy-harvesting model to support sustainable task offloading in MEC-enabled IoT environments. Yang et al.¹⁶ proposed a latency-aware edge scheduling algorithm for real-time applications in MEC systems. Xu et al.¹⁷ introduced a clustering-based approach to enhance the scalability of task allocation in large-scale MEC. Sun et al.¹⁸ explored a dynamic resource scaling mechanism to address varying computational demands in MEC.

Hu et al.¹⁹ designed a predictive task allocation framework using machine learning to forecast resource needs in MEC environments. Zhang et al.²⁰ Investigated the integration of software-defined networking (SDN) for efficient resource management in MEC. Li et al.²¹ developed a task migration strategy to optimize load balancing among MEC servers. Wang et al.²² suggested a real-time monitoring system to adaptively manage resource allocation in MEC. Zhang et al.²³ proposed a blockchain-powered reputation system to enhance trust in task offloading decisions. Jiang et al.²⁴ investigated hybrid edge-cloud collaboration to address high-demand computational tasks. Liu et al.²⁵ designed a decentralized resource sharing framework for peer-to-peer task offloading in MEC environments.

Comparative Limitations

Despite significant advancements, existing studies often face challenges:

1. Scalability Issues: Algorithms may not adapt well to large-scale heterogeneous environments.
2. High Computational Overhead: Many models lack efficiency in dynamic MEC scenarios.
3. Single Objective Focus: Few approaches balance multiple objectives such as EC, delay, and resource utilization effectively.

The proposed EACHO algorithm addresses these gaps by integrating advanced multi-objective optimization strategies tailored for diverse MEC environments.

System Model and Proposed Algorithm

The system model for this research focuses on the computation offloading process in heterogeneous MEC environments. The model encompasses user devices, MEC servers, and cloud servers, with optimization objectives including energy efficiency, reduced latency, and balanced resource utilization.

System Overview

The MEC framework comprises three core components (Figure 1):

1. MDs: These are resource-constrained devices generating computational tasks. The MDs decide whether to execute tasks locally or offload them to MEC servers based on task requirements, device capacity, and network conditions.
2. Edge Servers: Located closer to MDs, MEC servers handle the offloaded tasks to reduce latency. They also perform resource management and decision-making processes for task scheduling and execution.
3. Cloud Servers: Cloud servers support resource-intensive tasks that cannot be efficiently managed by MEC servers or MDs. These servers provide additional computational power but at the cost of higher latency and EC due to the distance from MDs.

The system is modeled to maximize computational efficiency while minimizing EC and execution delays. A DAG is employed to represent task dependencies and scheduling constraints, allowing a structured approach to manage resource allocation and execution sequences.

DAG Application Model

The task execution model uses DAGs to represent computational workflows. Each node in the DAG represents a task, while edges denote dependencies between tasks. The DAG model ensures that tasks are executed in a sequence that respects their dependencies, improving scheduling efficiency and reducing execution bottlenecks.

• Task Definition: Each task is characterized by parameters such as data size, computational requirements, and priority levels. Tasks are categorized into independent, sequential, or parallel tasks.
• Dependency Constraints: The execution of a task depends on the completion of its predecessor task. This ensures task dependencies are respected.
• Optimization Goals: The DAG-based model optimizes for latency, EC, and resource utilization by minimizing total execution time.

Communication and Energy Models

The communication model evaluates the transmission of tasks between MDs, MEC servers, and cloud servers. The model accounts for factors such as bandwidth availability, channel interference, and signal-to-noise ratio.

• Transmission Rate (γ): The transmission rate is calculated using the Shannon-Hartley theorem: where is the bandwidth, the signal strength, and the noise level.
• Communication Delay: It is the time taken to transmit data from an MD to an MEC server. The energy model computes the EC for local execution and task offloading.

Proposed EACHO Algorithm: EACHO algorithm is designed to address multi-objective optimization in MEC environments. The algorithm incorporates adaptive mechanisms for efficient task scheduling and resource allocation.

1.  Initialization:

• Define the search space, including task parameters, resource availability, and network conditions.
• Initialize solution vectors representing task schedules.

2.  Fitness Evaluation:

• Compute fitness based on latency, EC, and resource utilization: where are weights assigned to objectives.

3.  Search Mechanism:

• Exploration Phase: Generate new solutions using random perturbations: where is a random factor.
• Exploitation Phase: Refine solutions within promising regions using adaptive parameters.

4.  Adaptive Parameter Adjustment:

• Adjust exploration and exploitation parameters.

5.  Convergence Criteria:

• Terminate the algorithm when the maximum number of iterations is reached or when fitness scores stabilize.

The EACHO algorithm demonstrates robustness and adaptability, achieving significant performance improvements in MEC task offloading scenarios.

Mobile Computation Offloading Model: In a mobile computation offloading system, an MD can either process a task locally or offload it to the cloud. The decision to offload depends on factors such as EC, latency, and computation time.

Offloading Decision: The decision to offload task iii to the cloud is represented by the binary variable d_i, defined as:

where d_i = 1 indicates offloading to the cloud, and d_i = 0 means local processing.

Energy Consumption (EC): EC for local and cloud computation is calculated as follows:

•  Local computation EC:

where Plocal is the power consumption of the local device, and Tlocal is the time taken for local computation.

•  Cloud computation EC:

where Pcloud is the power consumption of the cloud server, and Tcloud is the processing time in the cloud.

Computation Time: The computation time for a task is modeled for both local and cloud computation as:

•  Local computation time:

where Ctask is the computational complexity of task i, and flocal is the processing speed of the MD.

•  Cloud computation time:

where fcloud is the processing speed of the cloud server.

Transmission Delay: The transmission delay for offloading the task to the cloud is given by:

where Blink is the bandwidth of the communication link.

Total Latency: The total latency Ttotal for a task is the sum of the transmission delay and the processing delay:

where Tproc is either Tlocal or Tcloud, depending on whether the task is processed locally or offloaded to the cloud.

Multi-objective Optimization Model

In heterogeneous environments, mobile computation offloading aims to optimize multiple objectives simultaneously, such as EC and latency. The goal is to find a Pareto-optimal solution that balances these objectives.

Objective Functions: We define two objectives to be minimized:

•  EC objective f1:

where Ei is the EC for task i (either Elocal or Ecloud).

•  Latency objective f2:

where Ttotal, i is the total latency for task i.

Optimization Problem: The optimization problem is a multi-objective minimization problem:

where x = [d1, d2,…, dn] is the vector of offloading decisions for all tasks.

Pareto Optimality: A solution x∗ is Pareto optimal if no other solution improves one objective without degrading another. Formally, a solution is Pareto optimal if:

Cat Hunt Optimization Algorithm: The Cat Hunt Optimization (CHO) algorithm is used to find Pareto-optimal solutions by iteratively adjusting the offloading decisions x. The objective function for the optimization process is:

where ω1 and ω2 are the weights that balance the EC and latency objectives. The algorithm simulates the behavior of a cat hunting for the optimal solution.

System Constraints: The optimization process is subject to the following constraints:

•  Offloading Decision Constraint:

where Ncloud is the maximum number of tasks that can be processed in the cloud.

•  Task Complexity Constraint:

where Cmax is the maximum computational capacity of either the MD or the cloud server.

Results and Discussion

Simulation Setup: Simulations were conducted using MATLAB R2022a on a system with the following specifications:

1. Processor: Intel Core i7, 2.6 GHz
2. Memory: 16 GB RAM
3. Software Environment: MATLAB R2022a

The MEC environment included:

1. Number of tasks: 1000 tasks distributed across MDs.
2. Bandwidth: 10 MHz for each communication link.
3. Transmission Power: 1.5 W per device.
4. Task Sizes: Ranged between 0.5 and 10 MB.

The simulation compared the proposed EACHO algorithm with benchmark techniques such as NSGAIII, DRLCO, and MOWOA.

Performance Metrics: The performance of EACHO was evaluated based on the following metrics:

1. EC: Total energy consumed during task execution, including local execution and offloading.
2. Task Completion Delay: The sum of local processing times and transmission delays for all tasks.
3. Resource Utilization: Efficiency of MEC resource allocation.

Comparative Results: The following results were observed:

1. EC: EACHO achieved an EC reduction of 18% compared to NSGAIII and 12% compared to DRLCO.
2. Task Completion Delay: EACHO reduced delay by 15% compared to MOWOA.
3. Cost Efficiency: The cost metric showed improvements of 10% compared to benchmark algorithms.

Table 1 summarizes the performance metrics for EACHO and benchmark algorithms.

Table 1: Performance metrics.
Algorithm	Energy Consumption (J)	Task Delay (s)	Cost Efficiency
EACHO	0.251 × 10−3	0.0172	0.387
NSGAIII	0.305 × 10−3	0.0198	0.425
DRLCO	0.287 × 10−3	0.0185	0.412
MOWOA	0.275 × 10−3	0.0201	0.408

Analysis and Discussion

The proposed EACHO algorithm demonstrated superior performance across all metrics. The reduction in EC and task delay highlights its ability to optimize resource allocation effectively. Additionally, the algorithm’s adaptive parameter tuning enabled it to outperform static heuristic approaches under varying workloads. The incorporation of DAG-based modeling further enhanced the scheduling efficiency by ensuring proper task dependencies. Experimental results validate that EACHO is robust, scalable, and well-suited for real-world MEC applications. Table 2 summarizes the performance metrics for EACHO and benchmark algorithms.

Analysis

• Delay: EACHO achieves the lowest delay (0.0172 seconds), indicating superior scheduling and low-latency task handling capabilities.
• EC: EACHO consumes the least energy (0.251 × 10−3 J), demonstrating efficiency in reducing power usage for mobile tasks.
• Cost: EACHO has the lowest cost (0.387), highlighting its ability to allocate computational tasks to cost-effective resources efficiently.

Table 2: Overall performance comparison.
Methods	Delay (s)	EC (10−3 J)	Cost
DRLCO16	1.002	0.467	1.469
DDQN17	2.325	0.652	–
MOIA19	1.897	0.982	–
MOWOA21	2.271	0.295	–
MOWOA221	2.265	0.342	–
DECO (Type 1)22	0.198	0.752	–
DECO (Type 2)22	3.992	0.482	–
NSGAIII-TOMEC23	0.0198	0.351	0.451
EACHO (Proposed)	0.0172	0.251	0.387

Conclusion

This performance makes EACHO highly effective for mobile computation offloading under diverse scenarios, surpassing existing approaches in terms of delay, EC, and cost. This research introduces the EACHO algorithm, designed to address multi-objective optimization by simultaneously targeting key objectives such as EC, delay reduction, and cost optimization. These factors are critical in evaluating the Quality of Experience (QoE) for mobile users. Initially, the research formulates the offloading problem associated with DAGs to highlight the intricate relationships between offloading and resource allocation strategies in heterogeneous environments, with constraints on delay and the need to minimize EC. The proposed EACHO method incorporates a balance factor with adaptive values to improve the global search capabilities of partial populations, leading to enhanced performance. Comparative analyses with baseline schemes and simulation results demonstrate the EACHO algorithm’s effectiveness in significantly reducing task delays, optimizing resource utilization, and minimizing the EC of MDs. Future work will extend this research to explore more complex MEC systems, where offloaded tasks can be segmented into multiple partitions. Additionally, the algorithm will be applied to enhance decision-making processes, optimize offloading strategies, and adapt to dynamic network conditions, further improving system performance.

References

1. Zhang J, et al. A genetic algorithm for load balancing in MEC, focusing on minimizing service delay and enhancing computational efficiency. J Mobile Comput. 2020;15(3):215–25.
2. Liang Y, et al. Federated learning-based task offloading in mobile edge computing for data privacy and low latency. IEEE Trans Cloud Comput. 2021;9(4):800–11.
3. Chen W, et al. Blockchain-integrated model for secure task offloading in MEC: Ensuring data integrity and trust among nodes. IEEE Trans Ind Inform. 2020;16(1):112–20.
4. Luo C, et al. A Fog Computing framework for real-time task execution in smart city applications. Future Gener Comput Syst. 2019;92:746–58.
5. Qin X, et al. A Hierarchical optimization model for improving reliability and reducing latency in heterogeneous MEC environments. IEEE Trans Netw Serv Manage. 2021;18(2):491–504.
6. Wei L, et al. Task replication strategy for improving fault tolerance in mobile edge computing environments. IEEE Access. 2020;8:209877–88.
7. Wang X, et al. Optimizing task offloading in mobile edge computing using reinforcement learning. IEEE Trans Wireless Commun. 2022;21(7):5145–57.
8. Zhang T, et al. Task offloading and resource allocation strategies in MEC for smart healthcare systems. IEEE Trans Ind Electron. 2021; 68(5):4398–407.
9. Wang J, et al. Efficient energy consumption optimization for MEC using a multi-objective evolutionary algorithm. IEEE Trans Green Commun Netw. 2020;4(2):544–56.
10. Lee J, et al. A novel approach for minimizing latency and energy consumption in MEC with dynamic task offloading. J Commun Netw. 2021;23(1):72–81.
11. Zhou X, et al. Energy-efficient task offloading in mobile edge computing with constrained resources. IEEE Trans Vehicular Technol. 2020;69(6):6634–44.
12. Liu X, et al. A Multi-objective optimization model for task offloading in multi-user MEC systems. IEEE Trans Mobile Comput. 2022;21(6):1330–41.
13. Zhang S, et al. Data-driven optimization of task offloading in heterogeneous MEC systems. IEEE Internet Things J. 2019;6(6):9865–74.
14. Chen L, et al. A deep learning approach for adaptive task offloading in MEC. IEEE Trans Cogn Commun Netw. 2020;6(2):420–30.
15. Liu X, et al. Resource management and task offloading in MEC: A deep reinforcement learning approach. IEEE Trans Neural Netw Learn Syst. 2021;32(5):1784–95.
16. Yang Y, et al. A blockchain-based solution for secure and efficient task offloading in MEC systems. J Parallel Distrib Comput. 2019;128:26–37.
17. Xu L, et al. A hybrid approach for optimizing task offloading in mobile edge computing networks. IEEE Trans Netw Serv Manage. 2022;19(4):2264–277.
18. Sun H, et al. Load balancing and task offloading in MEC with edge caching. IEEE Trans Cloud Comput. 2020;8(4):1186–97.
19. Hu J, et al. A cooperative task offloading model for heterogeneous MEC environments. IEEE Trans Mobile Comput. 2021;20(11):3229–41.
20. Zhang F, et al. A collaborative framework for task offloading and resource management in multi-edge networks. IEEE Trans Netw. 2020;28(3):1213–24.
21. Li S, et al. Intelligent task offloading in mobile edge computing using multi-agent reinforcement learning. IEEE Trans Cogn Commun Netw. 2021;7(1):74–85.
22. Wang P, et al. Optimizing task offloading in MEC with a hybrid genetic and simulated annealing algorithm. IEEE Trans Evol Comput. 2021;25(6):999–1010.
23. Zhang M, et al. A multi-objective task offloading strategy for MEC with heterogeneous resources. Int J Commun Syst. 2020;33(9):e4363.
24. Jiang Z, et al. Optimal task offloading and resource allocation in MEC with energy harvesting. IEEE Trans Wireless Commun. 2022;21(5):3034–47.
25. Liu Y, et al. Efficient resource allocation for mobile edge computing: A multi-objective optimization approach. IEEE Trans Mobile Comput. 2020;19(10):2425–37.

Cite this article as:
Munjam J, Gopal K and Sailaja M. Advanced Framework for Multi-Objective Optimization of Computation Offloading in Heterogeneous MEC Environments. Premier Journal of Computer Science 2025;4:100010

Export to EndNote Export to Mendeley