pith. sign in

arxiv: 2605.23152 · v1 · pith:CA3M63FGnew · submitted 2026-05-22 · 💻 cs.NI

Orchestrating Data Collection and Computation in Green IoT Networks

Junfei Zhan , Tengjiao He , Kwan-Wu Chin , Benyu Chen , Fei Song This is my paper

Pith reviewed 2026-05-25 03:13 UTC · model grok-4.3

classification 💻 cs.NI
keywords IoT networksenergy harvestingage of servicemixed integer linear programmingapplication schedulinggreen computingdata collection
0
0 comments X

The pith

A mixed integer linear program minimizes maximum age of service by jointly choosing sampling times, application runs, and energy use across energy-harvesting IoT nodes, gateways, and servers.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper introduces an optimization method for IoT networks where devices harvest energy to collect data and perform computations on servers. It formulates the first mixed integer linear program that decides device sampling intervals, whether to execute each application, and how energy is consumed at every network level. The objective is to minimize the highest age of service across all applications so that none waits too long. Two practical alternatives, a receding-horizon controller and a greedy embedding rule, are also developed and shown to produce maximum ages of service only 7 percent and 13 percent higher than the MILP solution.

Core claim

The central claim is that a mixed integer linear program can schedule and embed applications on energy-harvesting nodes by optimizing sampling times, binary decisions on running each application, and energy expenditures at devices, gateways, and servers, with the explicit goal of minimizing the maximum age of service across applications.

What carries the argument

The MILP that encodes sampling intervals, application-embedding binaries, and per-component energy variables while minimizing the maximum age of service.

If this is right

  • The MILP supplies an optimal benchmark against which any future scheduler for green IoT can be compared.
  • Receding-horizon control and greedy embedding become viable online substitutes that stay within 13 percent of the optimal maximum age of service.
  • Joint control of sampling rate, execution decisions, and energy draw keeps every application below a target age of service even when harvested power varies.
  • Energy accounting must span devices, gateways, and servers together; optimizing any subset in isolation leaves the global maximum age of service higher.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • If harvesting rates fluctuate faster than the MILP planning horizon, the performance gap between the optimal offline solution and the online heuristics may widen.
  • The same formulation could be re-used to enforce per-application latency bounds instead of a uniform maximum age of service by changing only the objective.
  • Adding a second objective that penalizes total energy draw would produce a Pareto frontier of schedules trading freshness against battery lifetime.

Load-bearing premise

The linear program is assumed to capture every relevant energy-harvesting dynamic, embedding constraint, and age-of-service rule without significant unmodeled effects from real hardware or wireless channels.

What would settle it

Compare the MILP-predicted maximum age of service against the age of service measured when the same schedule is executed on physical energy-harvesting nodes using recorded solar or RF traces.

Figures

Figures reproduced from arXiv: 2605.23152 by Benyu Chen, Fei Song, Junfei Zhan, Kwan-Wu Chin, Tengjiao He.

Figure 1
Figure 1. Figure 1: shows an example IoT network, where we aim to run two such applications. Their corresponding tasks, which are run as virtual network functions (VNFs), are deployed at gateways and at servers. VNF-Cs run on sensor devices. VNF￾Ps run on gateways or servers to process collected data, after which results are sent to the sink. Author Zhan is with the Jinan University-University of Birmingham Joint Institute, J… view at source ↗
Figure 2
Figure 2. Figure 2: An overview of RHCOP in a network with a single device a [PITH_FULL_IMAGE:figures/full_fig_p007_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Impact of various parameters on min-max AoS: (a) numb [PITH_FULL_IMAGE:figures/full_fig_p010_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Min-max AoS vs. average AoS. 1  [PITH_FULL_IMAGE:figures/full_fig_p011_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Runtime of MILP, RHCOP, GreedyOL, GMMPre and Random. [PITH_FULL_IMAGE:figures/full_fig_p011_5.png] view at source ↗
read the original abstract

Future Internet of things (IoT) networks will host applications that involve data collection and computation tasks on one or more servers. To this end, this paper proposes the first mixed integer linear program (MILP) to schedule and embed applications on energy harvesting nodes, where it optimizes (i) the sampling time of devices, (ii) whether to run an application, and (iii) the energy usage of devices, gateways and servers. To ensure applications are run often, we adopt the maximum age of service (AoS) metric, and set the MILP's objective to minimize the maximum AoS or min-max AoS of applications. This paper also proposes two novel solutions: (i) a receding horizon control (RHC) based method, and (ii) a solution that greedily embeds applications according to their AoS. The results show that the min-max AoS of RHC and greedy approach is respectively 1.07x and 1.13x higher than MILP.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 1 minor

Summary. The paper proposes the first mixed integer linear program (MILP) to schedule and embed applications on energy harvesting nodes in IoT networks. The MILP optimizes (i) the sampling time of devices, (ii) whether to run an application, and (iii) the energy usage of devices, gateways and servers, with the objective of minimizing the maximum age of service (AoS) across applications. It also introduces a receding horizon control (RHC) method and a greedy embedding heuristic, reporting that the min-max AoS achieved by RHC and greedy is respectively 1.07× and 1.13× higher than the MILP optimum.

Significance. If the MILP formulation is sound and the reported performance gaps are reproducible, the work could provide a useful optimization benchmark and practical heuristics for energy-aware scheduling in green IoT systems. The use of min-max AoS as the objective is a reasonable choice for ensuring fairness across applications.

major comments (2)
  1. [Abstract] Abstract: The claim that an MILP exists and the reported 1.07×/1.13× factors are presented without any description of decision variables, constraints, linearization techniques, or how the performance numbers were obtained from the solver output. This prevents verification that the formulation correctly encodes sampling times, application embedding, energy flows, and AoS semantics.
  2. [Results] Results section: No trace-driven validation, hardware experiment, or sensitivity analysis is supplied to confirm that the linear energy-harvesting model and AoS definition remain accurate under channel fading, battery non-linearities, or variable computation latency. The reported performance gaps are therefore load-bearing on an unvalidated modeling assumption.
minor comments (1)
  1. [Introduction] The assertion that this is the 'first' MILP would be strengthened by an explicit comparison table against prior MILP or optimization work on IoT energy harvesting in the introduction or related-work section.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the detailed review and the opportunity to clarify aspects of our work. We address the major comments point by point below.

read point-by-point responses

  1. Referee: [Abstract] Abstract: The claim that an MILP exists and the reported 1.07×/1.13× factors are presented without any description of decision variables, constraints, linearization techniques, or how the performance numbers were obtained from the solver output. This prevents verification that the formulation correctly encodes sampling times, application embedding, energy flows, and AoS semantics.

    Authors: The abstract is intended as a concise overview of the paper's contributions. The full MILP formulation is provided in Section 3 of the manuscript, where we define the decision variables (including binary variables for sampling decisions and application execution, continuous variables for energy consumption and AoS), the constraints modeling energy harvesting, storage, and consumption at devices, gateways, and servers, as well as the linearization of the min-max AoS objective. The performance results in Section 5 are obtained by solving the MILP using a commercial solver on the problem instances generated as described, and comparing the objective values achieved by the RHC and greedy methods to the MILP optimum. revision: no

  2. Referee: [Results] Results section: No trace-driven validation, hardware experiment, or sensitivity analysis is supplied to confirm that the linear energy-harvesting model and AoS definition remain accurate under channel fading, battery non-linearities, or variable computation latency. The reported performance gaps are therefore load-bearing on an unvalidated modeling assumption.

    Authors: The evaluation in the manuscript relies on simulation using linear energy harvesting models commonly adopted in the green IoT literature. We agree that additional validation would strengthen the results. In the revised manuscript, we will include a sensitivity analysis varying key parameters such as energy harvesting rates and computation latencies to assess the robustness of the performance gaps. However, hardware experiments and trace-driven validation with real channel fading data are beyond the scope of the current work and would require significant additional effort. revision: partial

Circularity Check

0 steps flagged

No circularity; MILP and heuristics compared via independent simulation outcomes

full rationale

The manuscript proposes an MILP that optimizes sampling times, application embedding decisions and energy flows under a min-max AoS objective, then reports simulation gaps (1.07× / 1.13×) versus RHC and greedy heuristics. No equations, fitted parameters renamed as predictions, or self-citation chains appear in the provided text. The performance numbers are presented as outputs of running the MILP versus the heuristics on the same model, not as definitions or tautologies. The derivation therefore remains self-contained; external validation questions (trace fidelity, hardware non-linearities) affect correctness risk but do not constitute circularity under the enumerated patterns.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract-only review provides no equations, no explicit modeling assumptions, and no parameter lists; therefore the ledger cannot be populated with concrete entries.

pith-pipeline@v0.9.0 · 5714 in / 1224 out tokens · 32860 ms · 2026-05-25T03:13:57.008543+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

33 extracted references · 33 canonical work pages

  1. [1]

    A rtificial intelligence-driven mechanism for edge computing-based i ndustrial ap- plications,

    A. H. Sodhro, S. Pirbhulal, and V . H. C. de Albuquerque, “A rtificial intelligence-driven mechanism for edge computing-based i ndustrial ap- plications,” IEEE Trans. on Ind. Inform. , vol. 15, no. 7, pp. 4235–4243, 2019

    work page 2019

  2. [2]

    VNF scheduling and sampling rat e max- imization in energy harvesting IoT networks,

    L. Zhang and K.-W. Chin, “VNF scheduling and sampling rat e max- imization in energy harvesting IoT networks,” IEEE Trans. on Mob. Comp., vol. 23, pp. 14441–14458, Dec. 2024

    work page 2024

  3. [3]

    Malware aw are UA V-assisted data collection and processing in solar-powe red IoT networks,

    T. He, M. Zheng, K.-W. Chin, T. Liu, and Y . Luo, “Malware aw are UA V-assisted data collection and processing in solar-powe red IoT networks,” IEEE Trans. Ind. Informat. , 2026. Early Access, doi: 10.1109/TII.2025.3644942

    work page doi:10.1109/tii.2025.3644942 2026

  4. [4]

    Trends in AI inference energy consumption: Beyond the performance -vs- parameter laws of deep learning,

    R. Desislavov, F. Mart´ ınez-Plumed, and J. Hern´ andez- Orallo, “Trends in AI inference energy consumption: Beyond the performance -vs- parameter laws of deep learning,” Sustainable Computing: Informatics and Systems , vol. 38, p. 100857, 2023

    work page 2023

  5. [5]

    A decade of video analytics at edge: Training, deployment, or chestration, and platforms,

    T. Bai, H. Zhao, L. Huang, Z. Wang, D. I. Kim, and A. Nallana than, “A decade of video analytics at edge: Training, deployment, or chestration, and platforms,” IEEE Commun. Surveys Tuts. , pp. 1–1, 2025

    work page 2025

  6. [6]

    Microservice dy namic migration based on age of service for edge computing,

    Y . Liu, B. Y ang, X. Y ang, Y . Wu, and C. Li, “Microservice dy namic migration based on age of service for edge computing,” in IEEE ICIT , (Shanghai, China), pp. 1–6, Aug. 2022

    work page 2022

  7. [7]

    Real-time status: Ho w often should one update?,

    S. Kaul, R. Y ates, and M. Gruteser, “Real-time status: Ho w often should one update?,” in IEEE INFOCOM , (Orlando, FL, USA), pp. 1–5, Mar. 2012

    work page 2012

  8. [8]

    Virtual ne twork embedding for wireless sensor networks time-efficient QoS/ QoI-aware approach,

    R. Katona, V . Cionca, D. O’Shea, and D. Pesch, “Virtual ne twork embedding for wireless sensor networks time-efficient QoS/ QoI-aware approach,” IEEE Internet Things J. , vol. 8, pp. 916–926, Jan. 2021

    work page 2021

  9. [9]

    Joi nt task scheduling and load balancing in computing power netwo rk- enabled edge computing systems,

    R. Xie, H. Zhu, Q. Tang, Q. Chen, S. Qiao, and T. Huang, “Joi nt task scheduling and load balancing in computing power netwo rk- enabled edge computing systems,” in 8th IEEE ICCC, (Chengdu, China), pp. 563–568, Dec. 2022

    work page 2022

  10. [10]

    Maximizing virtual ne twork embedding requests in RF-charging IoT networks,

    T. He, K. Chin, H. Ren, and X. Liu, “Maximizing virtual ne twork embedding requests in RF-charging IoT networks,” IEEE Commun. Lett., vol. 26, pp. 863–867, Apr. 2022

    work page 2022

  11. [11]

    Orchestrating virtual net work functions in wireless-powered IoT networks,

    H. Ren, K. W. Chin, and T. He, “Orchestrating virtual net work functions in wireless-powered IoT networks,” IEEE Internet Things J. , vol. 9, pp. 15874–15885, Sept. 2022

    work page 2022

  12. [12]

    Joint mul tioperator virtual network sharing and caching in energy harvesting-a ided environ- mental internet of things,

    Y . Cheng, J. Zhang, L. Y ang, C. Zhu, and H. Zhu, “Joint mul tioperator virtual network sharing and caching in energy harvesting-a ided environ- mental internet of things,” IEEE Internet Things J. , vol. 7, pp. 7689– 7701, Aug. 2020

    work page 2020

  13. [13]

    S ustainable virtual network function placement and traffic routing for g reen mobile edge networks,

    J. Liang, S. Huang, Y . Qiu, L. Liu, F. Aziz, and M. Chen, “S ustainable virtual network function placement and traffic routing for g reen mobile edge networks,” IEEE Trans. on Green Commun. and Netw. , vol. 4, pp. 1450–1465, Dec. 2024

    work page 2024

  14. [14]

    Age-based sche duling for monitoring and control applications in mobile edge computi ng systems,

    X. He, S. Wang, X. Wang, S. Xu, and J. Ren, “Age-based sche duling for monitoring and control applications in mobile edge computi ng systems,” in IEEE INFOCOM , (Virtual), pp. 1–10, May 2022

    work page 2022

  15. [15]

    AoI -aware incentive mechanism for mobile crowdsensing using stackel berg game,

    M. Xiao, Y . Xu, J. Zhou, J. Wu, S. Zhang, and J. Zheng, “AoI -aware incentive mechanism for mobile crowdsensing using stackel berg game,” in IEEE INFOCOM , (New Y ork, USA), pp. 1–10, May 2023

    work page 2023

  16. [16]

    Age-of-information-based computation offloadi ng and trans- mission scheduling in mobile-edge-computing-enabled IoT networks,

    Y . Jiang, J. Liu, I. Humar, M. Chen, S. A. AlQahtani, and M . S. Hossain, “Age-of-information-based computation offloadi ng and trans- mission scheduling in mobile-edge-computing-enabled IoT networks,” IEEE Internet Things J. , vol. 10, pp. 19782–19794, Nov. 2023

    work page 2023

  17. [17]

    Dynamic age minimization with real-time information preprocessing fo r edge-assisted IoT devices with energy harvesting,

    X. Ling, J. Gong, R. Li, S. Y u, Q. Ma, and X. Chen, “Dynamic age minimization with real-time information preprocessing fo r edge-assisted IoT devices with energy harvesting,” IEEE Trans. Netw. Sci. Eng., vol. 8, pp. 2288–2300, Sept. 2021

    work page 2021

  18. [18]

    Age of processed information minimization over fading multiple access channels,

    S. Jayanth and R. V . Bhat, “Age of processed information minimization over fading multiple access channels,” IEEE Trans. on Wirel. Commun. , vol. 22, pp. 1664–1676, Mar. 2023

    work page 2023

  19. [19]

    Enh ancing the safety of autonomous driving systems via AoI-optimized task scheduling,

    T. Shi, Q. Xu, J. Wang, C. Xu, K. Wu, K. Lu, and C. Qiao, “Enh ancing the safety of autonomous driving systems via AoI-optimized task scheduling,” IEEE Trans. V eh. Technol. , vol. 74, pp. 3804–3819, Mar. 2025

    work page 2025

  20. [20]

    Age of process ing: Age- driven status sampling and processing offloading for edge-c omputing- enabled real-time IoT applications,

    R. Li, Q. Ma, J. Gong, Z. Zhou, and X. Chen, “Age of process ing: Age- driven status sampling and processing offloading for edge-c omputing- enabled real-time IoT applications,” IEEE Internet Things J. , vol. 8, pp. 14471–14484, Oct. 2021

    work page 2021

  21. [21]

    Information freshness-aware task offloading in air-groun d integrated edge computing systems,

    X. Chen, C. Wu, T. Chen, Z. Liu, H. Zhang, M. Bennis, and Y . Ji, “Information freshness-aware task offloading in air-groun d integrated edge computing systems,” IEEE J. Sel. Areas Commun., vol. 40, pp. 243– 258, Jan. 2022

    work page 2022

  22. [22]

    Constrained federated learning for AoI-limited SFC in UA V-aided MEC for smart agriculture,

    M. Akbari, A. Syed, W. S. Kennedy, and M. Erol-Kantarci, “Constrained federated learning for AoI-limited SFC in UA V-aided MEC for smart agriculture,” IEEE Trans. on Machine Learning in Commun. and Netw. , vol. 1, pp. 277–295, Sept. 2023

    work page 2023

  23. [23]

    Age of information aware VNF schedul ing in industrial IoT using deep reinforcement learning,

    M. Akbari, M. R. Abedi, R. Joda, M. Pourghasemian, N. Mok ari, and M. Erol-Kantarci, “Age of information aware VNF schedul ing in industrial IoT using deep reinforcement learning,” IEEE J. Sel. Areas Commun., vol. 39, pp. 2487–2495, Aug. 2021

    work page 2021

  24. [24]

    AoI-aware par tial computation offloading in IIoT with edge computing: A deep re inforce- ment learning based approach,

    K. Peng, P . Xiao, S. Wang, and V . C. Leung, “AoI-aware par tial computation offloading in IIoT with edge computing: A deep re inforce- ment learning based approach,” IEEE Trans. Cloud Comput. , vol. 11, pp. 3766–3777, Oct. 2023

    work page 2023

  25. [25]

    Age-based scheduling fo r mobile edge computing: A deep reinforcement learning approach,

    X. He, C. Y ou, and T. Q. S. Quek, “Age-based scheduling fo r mobile edge computing: A deep reinforcement learning approach,” IEEE Trans. on Mob. Comp. , vol. 23, pp. 9881–9897, Oct. 2024

    work page 2024

  26. [26]

    Timely inf ormation updates for the internet of things with serverless computin g,

    S. Wakisaka, Y .-H. Chiang, H. Lin, and Y . Ji, “Timely inf ormation updates for the internet of things with serverless computin g,” in IEEE ICC, (Rome, Italy), pp. 1–6, June 2021

    work page 2021

  27. [27]

    Traffic eng ineering in software defined networks,

    S. Agarwal, M. Kodialam, and T. V . Lakshman, “Traffic eng ineering in software defined networks,” in IEEE INFOCOM , (Turin, Italy), pp. 1– 10, Apr. 2013

    work page 2013

  28. [28]

    Data-driven stochas tic models and policies for energy harvesting sensor communications,

    M.-L. Ku, Y . Chen, and K. J. R. Liu, “Data-driven stochas tic models and policies for energy harvesting sensor communications, ” IEEE J. Sel. Areas Commun. , vol. 33, pp. 1505–1520, Aug. 2015

    work page 2015

  29. [29]

    On the r ole of age of information in the Internet of things,

    M. A. Abd-Elmagid, N. Pappas, and H. S. Dhillon, “On the r ole of age of information in the Internet of things,” IEEE Commun. Mag. , vol. 5, pp. 72–77, Dec. 2019

    work page 2019

  30. [30]

    Recent advances on resource allocation in network function virtua lization,

    S. Y ang, F. Li, S. Trajanovski, R. Y ahyapour, and X. Fu, “ Recent advances on resource allocation in network function virtua lization,” IEEE Trans. Parallel Distrib. Syst. , vol. 32, pp. 295–314, Feb. 2021

    work page 2021

  31. [31]

    C. M. Bishop, Pattern Recognition and Machine Learning . Springer, 2006

    work page 2006

  32. [32]

    Deep-learnin g-assisted complete targets coverage in energy-harvesting IoT networ ks,

    K. Wang, C. Y ang, K.-W. Chin, and J. Xian, “Deep-learnin g-assisted complete targets coverage in energy-harvesting IoT networ ks,” IEEE Internet Things J. , vol. 12, pp. 17780–17790, June 2025

    work page 2025

  33. [33]

    Energy-savin g and resource-efficient algorithm for virtual network function placement with network scaling,

    M. Chen, Y . Sun, H. Hu, L. Tang, and B. Fan, “Energy-savin g and resource-efficient algorithm for virtual network function placement with network scaling,” IEEE Trans. Green Commun. Netw. , vol. 1, pp. 29–40, Mar. 2021

    work page 2021