The AI Native Digital Twin Architecture for Adaptive Resource Optimization Across the Edge Cloud Continuum

Dr Shashi kant; Vishnupriya B; Prabhavathi N

Authors

Dr Shashi kant st joseph university in tanzania Author
B.Vishnupriya Author
Prabhavathi N Author

Keywords:

Edge-cloud continuum, digital twin, reinforcement learning, task scheduling, heterogeneous computing, resource optimization.

Abstract

Latency-sensitive artificial-intelligence services increasingly span heterogeneous edge and cloud resources, where static schedulers struggle with bursty arrivals, variable network conditions, and competing performance objectives. This paper presents an AI-native digital-twin framework for adaptive task placement across an edge-cloud continuum. A synchronized digital twin maintains the observed resource state and supplies a predicted workload descriptor to a Deep Q-Network scheduler. The scheduler selects feasible task-to-node assignments while balancing latency, operating cost, energy consumption, resource utilization, and service-level agreement violations. The evaluation uses workload patterns derived from Alibaba Cluster Trace and Microsoft Azure virtual-machine traces in a discrete-time simulator containing heterogeneous CPU- and GPU-capable nodes. The proposed method is compared with First-Come-First-Served, Round Robin, a capacity-aware heuristic, and a standard DQN without digital-twin prediction. Based on the simulation values supplied for this study, the proposed method reduces mean latency by 44.0% and energy use by 40.0% relative to FCFS under high load, while increasing throughput by 40.0% and reducing the SLA-violation rate from 25% to 8%. These findings indicate that predictive state augmentation can improve scheduling decisions during demanding workload conditions. The conclusions remain limited to trace-driven simulation; deployment-scale validation and independently reproducible code are therefore identified as necessary next steps.

Downloads

Download data is not yet available.

Author Biographies

B.Vishnupriya

Latency-sensitive artificial-intelligence services increasingly span heterogeneous edge and cloud resources, where static schedulers struggle with bursty arrivals, variable network conditions, and competing performance objectives. This paper presents an AI-native digital-twin framework for adaptive task placement across an edge-cloud continuum. A synchronized digital twin maintains the observed resource state and supplies a predicted workload descriptor to a Deep Q-Network scheduler. The scheduler selects feasible task-to-node assignments while balancing latency, operating cost, energy consumption, resource utilization, and service-level agreement violations. The evaluation uses workload patterns derived from Alibaba Cluster Trace and Microsoft Azure virtual-machine traces in a discrete-time simulator containing heterogeneous CPU- and GPU-capable nodes. The proposed method is compared with First-Come-First-Served, Round Robin, a capacity-aware heuristic, and a standard DQN without digital-twin prediction. Based on the simulation values supplied for this study, the proposed method reduces mean latency by 44.0% and energy use by 40.0% relative to FCFS under high load, while increasing throughput by 40.0% and reducing the SLA-violation rate from 25% to 8%. These findings indicate that predictive state augmentation can improve scheduling decisions during demanding workload conditions. The conclusions remain limited to trace-driven simulation; deployment-scale validation and independently reproducible code are therefore identified as necessary next steps.
Prabhavathi N

Latency-sensitive artificial-intelligence services increasingly span heterogeneous edge and cloud resources, where static schedulers struggle with bursty arrivals, variable network conditions, and competing performance objectives. This paper presents an AI-native digital-twin framework for adaptive task placement across an edge-cloud continuum. A synchronized digital twin maintains the observed resource state and supplies a predicted workload descriptor to a Deep Q-Network scheduler. The scheduler selects feasible task-to-node assignments while balancing latency, operating cost, energy consumption, resource utilization, and service-level agreement violations. The evaluation uses workload patterns derived from Alibaba Cluster Trace and Microsoft Azure virtual-machine traces in a discrete-time simulator containing heterogeneous CPU- and GPU-capable nodes. The proposed method is compared with First-Come-First-Served, Round Robin, a capacity-aware heuristic, and a standard DQN without digital-twin prediction. Based on the simulation values supplied for this study, the proposed method reduces mean latency by 44.0% and energy use by 40.0% relative to FCFS under high load, while increasing throughput by 40.0% and reducing the SLA-violation rate from 25% to 8%. These findings indicate that predictive state augmentation can improve scheduling decisions during demanding workload conditions. The conclusions remain limited to trace-driven simulation; deployment-scale validation and independently reproducible code are therefore identified as necessary next steps.

References

[1] V. Mnih et al., “Human-level control through deep reinforcement learning,” Nature, vol. 518, no. 7540, pp. 529–533, 2015.

[2] R. S. Sutton and A. G. Barto, Reinforcement Learning: An Introduction, 2nd ed. Cambridge, MA, USA: MIT Press, 2018.

[3] C. J. C. H. Watkins and P. Dayan, “Q-learning,” Machine Learning, vol. 8, pp. 279–292, 1992.

[4] H. van Hasselt, A. Guez, and D. Silver, “Deep reinforcement learning with double Q-learning,” in Proc. AAAI Conf. Artificial Intelligence, 2016.

[5] M. G. Bellemare, W. Dabney, and R. Munos, “A distributional perspective on reinforcement learning,” in Proc. ICML, 2017.

[6] M. Satyanarayanan, “The emergence of edge computing,” Computer, vol. 50, no. 1, pp. 30–39, 2017.

[7] W. Shi et al., “Edge computing: Vision and challenges,” IEEE Internet of Things Journal, vol. 3, no. 5, pp. 637–646, 2016.

[8] S. Deng et al., “Edge intelligence: The confluence of edge computing and artificial intelligence,” IEEE Internet of Things Journal, vol. 7, no. 8, pp. 7457–7469, 2020.

[9] H. Gupta et al., “iFogSim: A toolkit for modeling and simulation of resource management techniques in the Internet of Things, edge and fog computing environments,” Software: Practice and Experience, vol. 47, no. 9, pp. 1275–1296, 2017.

[10] G. Premsankar, M. Di Francesco, and T. Taleb, “Edge computing for the Internet of Things: A case study,” IEEE Internet of Things Journal, vol. 5, no. 2, pp. 1275–1284, 2018.

[11] Z. Wang et al., “Mobility digital twin: Concept, architecture, and applications,” IEEE Internet of Things Journal, 2022.

[12] F. Tao and M. Zhang, “Digital twin shop-floor: A new shop-floor paradigm,” Journal of Manufacturing Systems, vol. 48, pp. 237–249, 2019.

[13] A. Fuller, Z. Fan, C. Day, and C. Barlow, “Digital twin: Enabling technologies, challenges and open research,” IEEE Access, vol. 8, pp. 108952–108971, 2020.

[14]–[30] Entries retained from the source manuscript require bibliographic verification and completion before submission.

[15] K. Deb, Multi-Objective Optimization Using Evolutionary Algorithms. Chichester, U.K.: Wiley, 2001.

[32]–[34] Entries retained from the source manuscript require bibliographic verification and completion before submission.

[16] R. N. Calheiros et al., “CloudSim: A toolkit for modeling and simulation of cloud computing environments and evaluation of resource provisioning algorithms,” Future Generation Computer Systems, vol. 27, no. 1, pp. 23–50, 2011.

[17] Alibaba Group, “Alibaba cluster trace program,” public workload trace dataset, 2018.

[18] Microsoft, “Azure public dataset,” virtual-machine workload traces, 2020.

[19]–[40] Entries retained from the source manuscript require bibliographic verification and completion before submission.