Optimization of Mobile Edge Computing and Networking for Ultra-Low Latency and Green Internet of Things Applications

Abstract

IoT growth has quietly outpaced what cloud-centric network architectures were built to handle. When billions of sensors, actuators, and embedded controllers started producing continuous data streams, the assumption that round-trip latency toadistantserverwasanacceptablecoststoppedholding.In industrial automation lines, hospital monitoring loops, and vehicle-to-vehicle coordination, tens of milliseconds separate a correctly functioning system from a failed one. This paper examines Mobile Edge Computing (MEC) as a structural fix: computation moves physically closer to wheredataoriginates,cuttingpropagationdelayandrelievingbackhaul pressureatthesametime.Welookathowtaskoffloading,energy budgeting, and shared resource allocation actually behave under field conditions — varying wireless channels, heterogeneous workloads, energy supplies that aren’t always predictable. The optimization framework we propose uses a Deep Q-Network (DQN) that learns from what the system does rather than from a pre-built model of what it’s supposed to do, which matters when operating conditions drift. One finding we want to highlight upfront: minimizing latency and minimizing energy are not genuinely competing objectives.A well-trained edge policy can satisfy timing requirements while simultaneously cutting power draw across the device, network, and server layers.

Introduction

Modern hospital and IoT environments require highly different latency levels depending on the application. While routine monitoring can tolerate delays, critical systems like ventilators and emergency alerts need near-instant responses (around 10 ms or less). Traditional cloud computing struggles with these strict deadlines due to unavoidable physical latency from long-distance data transmission. This limitation motivates Mobile Edge Computing (MEC), where computation is moved closer to devices by placing servers at or near base stations, reducing both delay and network load.

MEC operates through a layered architecture consisting of edge servers, MEC hosts, and a software platform that manages task scheduling and resource allocation. When a task arrives, it is either processed locally at the edge for ultra-low latency or forwarded to the cloud if it is too complex. This improves response time and also reduces energy consumption by limiting data transmission over long network paths.

Energy efficiency is a key advantage of MEC. Processing data at the edge reduces backhaul traffic, lowering power usage across devices, network infrastructure, and data centers. Techniques like adaptive offloading, server power scaling, and energy-aware scheduling help optimize performance, especially in large IoT deployments and energy-harvesting devices.

Latency in MEC systems comes from multiple sources: wireless transmission delays, server queuing, computation time, and result transmission. Among these, wireless and queueing delays are often the most significant. Optimizing MEC requires balancing these factors rather than focusing on computation alone, since improving one part (e.g., CPU speed) may worsen others (e.g., thermal throttling and queue buildup).

Emerging research directions include split inference (dividing neural networks between devices and edge), federated learning (training models without centralizing data), UAV-based mobile edge servers for flexible coverage, and reconfigurable intelligent surfaces (RIS) to improve wireless connectivity.

Conclusion

MEC works. The latency and energy reductions it produces are not theoretical projections — they follow from moving computationphysicallyclosertowheredataisgenerated, andthemechanismiswellunderstood.What’sharderto close is the gap between what MEC can do in a well- configureddeploymentandwhatitconsistentlydeliversunder the messy conditions of real infrastructure: dynamic traffic, heterogeneous devices, shifting channel statistics, and energy supplies that don’t hold to schedule. Threeproblemssitatthecenterofthatgap.First,offloading policiesthatperformwellincontrolledsettingstendtodegrade as operating conditions drift, because most of them embed assumptions about the environment that stop holding over time.Second,multi-tenantresourceschedulingonsharededgeserversrequireshandlinggenuineworkloadheterogeneity—not averaged abstractions — to avoid systematic unfairness or priority inversions at peak load. Third, energy management across device, network, and server tiers needs to be treated asa coupled problem, not a per-tier optimization. The DQN-based framework described here targets the of- floadingpiece.Itlearnsfromexperienceratherthanfroma pre-calibrated model, which gives it the ability to track shifting conditions without manual re-tuning. On its own it doesn’t solve the multi-tenant scheduling or cross-tier energy coupling problems, but it provides a foundation that can be extended.Plannednextstepsincludemulti-agentcoordination to handle competition among concurrent offloading decisions, integration of energy harvesting forecasts into the reward signal, and evaluation against mobility traces that include realistic handover sequences.

References

[1] K.Zhang,Y.Mao,S.Leng,Q.Zhao,L.Zhang,X.Peng,L.Pan, [2] S.Maharjan,andY.Zhang,“Energy-efficientoffloadingformobileedgecomputingin5Gheterogeneousnetworks,”IEEEAccess,vol.4,pp.5896–5907,2016. [3] Z.Ren,Z.Liu,andY.Chen,“Ultra-lowlatencyedgecomputingforimmersive AR/VR applications,” IEEE Trans. Mobile Comput., 2025. [4] S.-Q. Lee, D.-H. Han, and J.-H. Kim, “URLLC-aware routing forindustrial IoT in 5G networks,” in Proc. IEEE VTC, 2020. [5] R. Dhumpati, A. Sharma, and K. Rao, “Adaptive resource allocation forheterogeneous MEC environments,” IEEE Internet Things J., 2025. [6] M. Ergen, B. Aydin, and T. Yilmaz, “Beyond 5G edge computing:Architectural considerations for 6G,” IEEE Commun. Mag., 2024. [7] T. V. Thai, N. H. Tran, and C. S. Hong, “MIMO-assisted mobile edgecomputing:Asurveyfor6Gsystems,”IEEECommun.Surv.Tut.,2023. [8] A. Almalawi, X. Yu, Z. Tari, A. Fahad, and I. Khalil, “Intelligent edgecachingforlatency-sensitiveIoTservices,”IEEETrans.ParallelDistrib.Syst., 2023. [9] S.Taimoor,M.Ali,andA.Khan,“UAV-assistedmobileedgecomputing:Energyandcoverageoptimization,”IEEEWirelessCommun.Lett.,2023. [10] Z. Trabelsi, H. Afifi, and F. Zarai, “Fuzzy logic-based task offloadingfor Internet of Vehicles,” in Proc. IEEE ICC, 2023. [11] L. Toka, “Kubernetes-based orchestration for URLLC edge applica-tions,” in Proc. IEEE INFOCOM, 2022. [12] L.Baresi,D.Filgueira,andG.Quattrocchi,“Serverlessedgecomputing:Challenges and opportunities,” IEEE Pervasive Comput., vol. 21, no. 2,pp.48–56,2022. [13] S. Banoth and R. Thakur, “Survey of edge computing architectures forreal-time IoT data processing,” IEEE Access, 2023. [14] A. K. Mishra, “Joint latency and throughput optimization in multi-tieredge computing networks,” IEEE Trans. Netw. Sci. Eng., 2023. [15] P. Mach and Z. Becvar, “Mobile edge computing: A survey on archi-tectureandcomputationoffloading,”IEEECommun.Surv.Tut.,vol.19,no. 3, pp. 1628–1656, 2017. [16] X. Chen, H. Zhang, C. Wu, S. Mao, Y. Ji, and M. Bennis, “Optimizedcomputation offloading performance in virtual edge computing systemsvia deep reinforcement learning,” IEEE Internet Things J., vol. 6, no. 3,pp.4005–4018,2019.

Copyright

Copyright © 2026 Prasidhee Koli, Darshan Bhangale, Dhrumi Karia, Dhruv Chawda, Mayur Hingu, Prithvi Bhadra, Neha Panadi, Dr. Asha Durafe. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.