arxiv: 2605.04023 · v1 · submitted 2026-05-05 · 💻 cs.GT · cs.PF

Recognition: unknown

Decentralized Edge Caching under Budget and Storage Constraints: A Game-Theoretic Approach

Hamta Sedghani , Zahra Seyedi , Mauro Passacantando , Danilo Ardagna

Authors on Pith no claims yet

Pith reviewed 2026-05-07 03:59 UTC · model grok-4.3

classification 💻 cs.GT cs.PF

keywords edge cachinggame theorypotential gameStackelberg gamecontent providersNash equilibriumdecentralized resource allocationstorage constraints

0 comments

The pith

Content provider competition for edge storage forms an exact potential game under light constraints, guaranteeing decentralized Nash equilibria.

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

The paper models the allocation of edge-device storage among competing content providers as a hierarchical game that combines a Stackelberg interaction between each provider and the devices with a non-cooperative game among the providers themselves. It shows that when storage limits are mild this competition is an exact potential game, which guarantees a pure-strategy Nash equilibrium and allows simple decentralized best-response updates to reach it. Numerical experiments indicate that convergence remains stable even after storage constraints bind, but the resulting allocations exhibit wider profit gaps among providers and greater leverage for the edge devices. The framework therefore supplies a budget-respecting way to allocate scarce caching capacity without a central authority.

Core claim

The authors establish that the non-cooperative game played by content providers constitutes an exact potential game whenever storage constraints remain light. This structure ensures the existence of a pure-strategy Nash equilibrium that can be reached by decentralized best-response dynamics. When storage constraints become binding the potential-game property is lost, yet extensive numerical experiments show that convergence remains stable in practice and that scarcity increases inequality in provider profits while shifting bargaining power toward the edge devices.

What carries the argument

Exact potential game among content providers under light storage constraints, embedded inside a Stackelberg game between providers and edge devices.

If this is right

Pure-strategy Nash equilibria exist for content-provider competition when storage limits are mild.
Decentralized best-response dynamics converge to equilibrium without central coordination.
Binding storage constraints produce greater inequality in profits across content providers.
Edge devices obtain increased relative bargaining power as storage becomes scarce.
Convergence speed and stability depend more on the intensity of provider competition than on the number of edge devices.

Where Pith is reading between the lines

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

If the exact-potential structure extends to other budgeted resource markets, similar decentralized algorithms could coordinate cloud or network-slicing allocations without heavy central oversight.
The documented rise in provider inequality under scarcity suggests that subsidies or access rules could be calibrated to the same model parameters to reduce disparities.
Replacing the simulation cost and budget distributions with empirical traces from live mobile networks would test whether the observed convergence persists outside the chosen parameter ranges.

Load-bearing premise

The interactions between content providers and edge devices can be accurately represented by a Stackelberg leader-follower structure and that the chosen simulation parameters capture real-world heterogeneity in budgets, costs, and storage.

What would settle it

A deployment of edge caching with multiple competing providers under light storage limits in which the providers' storage allocations fail to converge to any stable point, or in which measured profit distributions remain equal rather than widening under storage scarcity.

Figures

Figures reproduced from arXiv: 2605.04023 by Danilo Ardagna, Hamta Sedghani, Mauro Passacantando, Zahra Seyedi.

**Figure 1.** Figure 1: Network Architecture all notations used in the paper are summarized in Table II in Appendix A. A. Network Architecture The network architecture of the secure edge caching model is depicted in view at source ↗

**Figure 2.** Figure 2: Average iterations and execution time to converge vs. number of view at source ↗

**Figure 3.** Figure 3: Average iterations and execution time to converge vs. number of view at source ↗

**Figure 4.** Figure 4: Average CP utilities vs. ED cost parameter view at source ↗

**Figure 5.** Figure 5: Average ED utility vs. ED cost parameter view at source ↗

read the original abstract

The rapid growth of mobile social networks (MSNs) has significantly increased the demand for low-latency and reliable content delivery, motivating the deployment of edge caching systems. In practice, multiple content providers (CPs) compete for the limited storage resources of edge devices (EDs), while facing heterogeneous budgets and operational costs. This paper investigates a decentralized multi-CP edge caching framework that jointly accounts for CP budget constraints, ED storage limitations, and strategic interactions among all entities. We formulate the interaction between CPs and EDs as a hierarchical game, combining a Stackelberg model for CP-ED interactions with a non-cooperative game among competing CPs. Under light storage constraints, we show that CP competition constitutes an exact potential game, ensuring the existence of a pure-strategy Nash equilibrium and enabling decentralized convergence. When storage constraints are binding, the resulting game loses this structure; nevertheless, extensive simulations demonstrate stable and efficient convergence in practice. Through a comprehensive numerical evaluation, we show that convergence behavior is primarily driven by CP competition rather than the scale of edge infrastructure. We further reveal that storage scarcity fundamentally alters economic outcomes, amplifying inequality among CPs while increasing the relative bargaining power of EDs. The proposed framework provides a scalable and economically grounded solution for decentralized resource allocation in multi-provider edge caching systems.

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.

Desk Editor's Note private letter to a colleague

Applies standard Stackelberg and potential-game tools to multi-CP edge caching with budgets and storage limits, but the composed utilities after ED best responses need explicit verification to support the exact potential claim.

read the letter

Dear Colleague, The main takeaway is a hierarchical model for content providers competing over edge-device storage under budget and storage constraints, with a Stackelberg layer between CPs and EDs plus a non-cooperative game among CPs. Under light constraints they show the CP game is an exact potential game, guaranteeing a pure Nash equilibrium and decentralized convergence; simulations then handle the binding case and report effects like greater inequality among CPs and stronger ED bargaining power when storage is scarce. The observation that convergence is driven more by CP competition than by the number of edge nodes is a usable practical note. The formulation adds heterogeneous budgets and costs to earlier caching games, which is a reasonable incremental step. The work is an application of known results rather than a new theoretical device. The soft spot is the potential-game step itself. Each CP's effective payoff is the value after the EDs play their best response to the full vector of proposals, so the utility is a composition. The paper appears to construct the potential on the original payoffs and asserts the property survives, but without checking that the best-response mapping preserves the required cross-partial cancellation, the exact-potential guarantee is not fully secured. Simulation parameters and the precise handling of binding constraints are also light on detail, which limits how far one can trust the reported outcomes. This paper is for people working on game-theoretic resource allocation in edge networks. A reader already interested in decentralized caching models would pick up the inequality results and the convergence behavior. It is not broad-impact work, but the model is concrete enough and the claims are checkable, so it merits referee time. I would send it for peer review and ask the authors to spell out the composed potential function and to expand the simulation setup.

Referee Report

1 major / 2 minor

Summary. The manuscript presents a hierarchical game-theoretic model for decentralized multi-CP edge caching, where CPs compete for ED storage under heterogeneous budgets and costs. CP-ED interactions are modeled as a Stackelberg game, while CP competition is a non-cooperative game. Under light (non-binding) storage constraints, the authors claim that the CP game is an exact potential game, guaranteeing a pure-strategy Nash equilibrium and decentralized convergence via best-response dynamics. When storage constraints bind, the potential-game structure is lost, but simulations are used to show practical convergence and to analyze how scarcity increases CP inequality and ED bargaining power. The evaluation concludes that convergence is driven primarily by CP competition rather than edge infrastructure scale.

Significance. If the central claim that the composed CP game remains an exact potential game under light constraints holds, the work would provide a scalable, economically grounded decentralized allocation mechanism for edge caching in mobile social networks. The potential-game result would deliver convergence guarantees without central coordination, while the simulation analysis of binding constraints and economic outcomes (inequality, bargaining power) would link technical limits to fairness considerations. The separation of light-constraint theory from binding-constraint empirics is a reasonable scope choice. However, the significance is conditional on verification that the Stackelberg composition preserves the potential property.

major comments (1)

[Theoretical analysis of CP game under light constraints] § on CP potential game under light constraints (exact location not numbered in abstract but central to the theoretical result): The potential function is constructed on the pre-composition CP utilities. The manuscript does not explicitly verify that the composition with the ED best-response mapping BR_ED(s) preserves the exact-potential cross-partial cancellation condition. Because each CP's effective utility is u_i(s_i, s_{-i}) = f_i(s_i, BR_ED(s)), the cross derivatives after substitution must still satisfy the potential condition; this step is load-bearing for the existence of a pure-strategy NE and for the decentralized convergence claim.

minor comments (2)

[Numerical evaluation] Simulation section: the specific parameter values, ranges for heterogeneous budgets, storage limits, and cost functions are not detailed in the abstract; providing them (or a reproducibility appendix) would strengthen the claim that the parameters sufficiently represent real-world heterogeneity when constraints bind.
[Model formulation] Notation: the distinction between light and binding storage constraints should be formalized with an explicit threshold or condition on the storage vector to avoid ambiguity when the potential-game property ceases to hold.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for the thorough and constructive review of our manuscript. The major comment regarding the need for explicit verification that the exact potential property is preserved after composing the CP utilities with the ED best-response mapping is well-taken and points to a step that merits additional detail. We will revise the paper to address this directly. Our point-by-point response is provided below.

read point-by-point responses

Referee: The potential function is constructed on the pre-composition CP utilities. The manuscript does not explicitly verify that the composition with the ED best-response mapping BR_ED(s) preserves the exact-potential cross-partial cancellation condition. Because each CP's effective utility is u_i(s_i, s_{-i}) = f_i(s_i, BR_ED(s)), the cross derivatives after substitution must still satisfy the potential condition; this step is load-bearing for the existence of a pure-strategy NE and for the decentralized convergence claim.

Authors: We appreciate the referee's precise identification of this subtlety. The original manuscript constructs the potential function for the CP utilities f_i prior to the Stackelberg composition with the EDs. We agree that an explicit check confirming the composed utilities u_i(s) = f_i(s_i, BR_ED(s)) continue to satisfy the exact-potential condition (via cross-partial cancellation) is necessary and was not detailed. In the revised manuscript we will insert a dedicated lemma that derives the effective utilities under light (non-binding) storage constraints, computes the relevant partial derivatives after substitution of BR_ED(s), and verifies that the cross-partial symmetry required for an exact potential function still holds. This addition will rigorously support the existence of a pure-strategy Nash equilibrium and the convergence guarantees for best-response dynamics. We thank the referee for helping us strengthen the theoretical foundation. revision: yes

Circularity Check

0 steps flagged

No significant circularity; potential-game claim rests on standard construction from utilities

full rationale

The central result states that under light storage constraints the CP game is an exact potential game. This is obtained by exhibiting a potential function whose differences recover the payoff differences of the composed utilities; the construction is the standard one from potential-game theory and does not reduce to a fitted parameter, a self-referential definition, or a prior result by the same authors. No equations in the provided text equate the equilibrium existence claim to its own inputs by construction. Simulations are presented separately and are not used to define the theoretical property. The derivation chain therefore remains self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

2 free parameters · 2 axioms · 0 invented entities

The model relies on standard game-theoretic axioms and treats budgets, operational costs, and storage capacities as exogenous inputs rather than deriving them. No new entities are postulated.

free parameters (2)

heterogeneous budgets
Budgets are given inputs to the model and not derived from first principles.
storage limits
ED storage capacities are treated as fixed constraints.

axioms (2)

standard math Existence of pure-strategy Nash equilibrium in exact potential games
Invoked to guarantee equilibrium under light storage constraints.
domain assumption Stackelberg leader-follower structure for CP-ED interactions
Assumes the hierarchical interaction follows the Stackelberg model without additional justification in the abstract.

pith-pipeline@v0.9.0 · 5542 in / 1467 out tokens · 61088 ms · 2026-05-07T03:59:41.733669+00:00 · methodology

Review history (2 revisions) →

discussion (0)

Reference graph

Works this paper leans on

32 extracted references

[1]

Cost-effective edge data caching with failure tolerance and popularity awareness,

R. Luo et al., “Cost-effective edge data caching with failure tolerance and popularity awareness,”IEEE Trans. on Mobile Computing, 2025. [2]Https://www.ericsson.com/en/reports-and-papers/mobility-report /dataforecasts/mobile-traffic-forecast

2025
[2]

A survey on privacy-preserving caching at network edge: Classification, solutions, and challenges,

X. Zhang et al., “A survey on privacy-preserving caching at network edge: Classification, solutions, and challenges,”ACM Computing Sur- veys, vol. 57, no. 5, pp. 1–38, 2025

2025
[3]

Dynamic task offloading and resource allocation for noma-aided mobile edge computing: An energy efficient design,

Y . Chen et al., “Dynamic task offloading and resource allocation for noma-aided mobile edge computing: An energy efficient design,”IEEE Trans. on Services Computing, vol. 17, no. 4, pp. 1492–1503, 2024

2024
[4]

Game theory and reinforcement learning based secure edge caching in mobile social networks,

Q. Xu et al., “Game theory and reinforcement learning based secure edge caching in mobile social networks,”IEEE Trans. on Information Forensics and Security, vol. 15, pp. 3415–3429, 2020

2020
[5]

A survey of ddos attack and defense technologies in multi-access edge computing,

Y . Ma et al., “A survey of ddos attack and defense technologies in multi-access edge computing,”IEEE Internet of Things Journal, 2024

2024
[6]

Design and implementation of social based edge node selection algorithm,

Q. He et al., “Design and implementation of social based edge node selection algorithm,”Multimedia Tools and Applications, vol. 83, no. 34, pp. 81 127–81 149, 2024

2024
[7]

Auction theory and game theory based pricing of edge computing resources: A survey,

J. Yu et al., “Auction theory and game theory based pricing of edge computing resources: A survey,”IEEE Internet of Things Journal, 2025

2025
[8]

Content caching in mobile edge computing: A survey,

Y . Khan et al., “Content caching in mobile edge computing: A survey,” Cluster Computing, vol. 27, no. 7, pp. 8817–8864, 2024

2024
[9]

Context-aware proactive edge caching for vehicular edge computing based on asynchronous federated learning,

Z. Liao et al., “Context-aware proactive edge caching for vehicular edge computing based on asynchronous federated learning,”IEEE Internet of Things Journal, 2025

2025
[10]

An incentive mechanism based on a stackelberg game for mobile crowdsensing systems with budget constraint,

H. Sedghani et al., “An incentive mechanism based on a stackelberg game for mobile crowdsensing systems with budget constraint,”Ad Hoc Networks, vol. 123, p. 102 626, 2021

2021
[11]

Federated deep reinforcement learning for multimodal content caching in edge-cloud networks,

W. Feng et al., “Federated deep reinforcement learning for multimodal content caching in edge-cloud networks,”IEEE Trans. on Network Science and Engineering, 2025

2025
[12]

Attention-enhanced prioritized proximal policy optimization for adaptive edge caching,

F. Niknia et al., “Attention-enhanced prioritized proximal policy optimization for adaptive edge caching,”IEEE Trans. on Vehicular Technology, 2025

2025
[13]

Pencache: Popularity based cooperative caching in named data networks,

P. Chaudhary et al., “Pencache: Popularity based cooperative caching in named data networks,”Computer Networks, vol. 257, p. 110 995, 2025

2025
[14]

Optimal multi-bitrate video caching and processing in edge computing: A stackelberg game approach,

M. Guo et al., “Optimal multi-bitrate video caching and processing in edge computing: A stackelberg game approach,”IEEE Internet of Things Journal, 2025

2025
[15]

Pricing-driven service caching and task offloading in mobile edge computing,

J. Yan et al., “Pricing-driven service caching and task offloading in mobile edge computing,”IEEE Trans. on Wireless Communications, vol. 20, no. 7, pp. 4495–4512, 2021

2021
[16]

Dynamic clustering for low-delay delivery of video content cached in mec servers,

A. Doostmohammadi et al., “Dynamic clustering for low-delay delivery of video content cached in mec servers,”IEEE Systems Journal, vol. 17, no. 4, pp. 5842–5853, 2023

2023
[17]

Optimal push and pull-based edge caching for dynamic content,

B. Abolhassani et al., “Optimal push and pull-based edge caching for dynamic content,”IEEE/ACM Trans. on Networking, vol. 32, no. 4, pp. 2765–2777, 2024

2024
[18]

Investment-driven budget allocation and dynamic pricing strategies in edge cache network,

Q. Wang et al., “Investment-driven budget allocation and dynamic pricing strategies in edge cache network,”Pervasive and Mobile Computing, vol. 109, p. 102 040, 2025

2025
[19]

Efficient online computing offloading for budget- constrained cloud-edge collaborative video streaming systems,

S. Yuan et al., “Efficient online computing offloading for budget- constrained cloud-edge collaborative video streaming systems,”IEEE Trans. on Cloud Computing, 2024

2024
[20]

A survey on resource scheduling approaches in multi-access edge computing environment: A deep reinforcement learning study,

A. A. Ismail et al., “A survey on resource scheduling approaches in multi-access edge computing environment: A deep reinforcement learning study,”Cluster Computing, vol. 28, no. 3, p. 184, 2025

2025
[21]

A stackelberg game based framework for edge pricing and resource allocation in mobile edge computing,

S. Cheng et al., “A stackelberg game based framework for edge pricing and resource allocation in mobile edge computing,”IEEE Internet of Things Journal, 2024

2024
[22]

Contract theory and stackelberg game based storage resource allocation in edge caching systems,

Y . Fan et al., “Contract theory and stackelberg game based storage resource allocation in edge caching systems,”IEEE Internet of Things Journal, 2024

2024
[23]

A game-theoretic analysis of joint mobile edge caching and peer content sharing,

C. Jiang et al., “A game-theoretic analysis of joint mobile edge caching and peer content sharing,”IEEE Trans. on Network Science and Engineering, vol. 10, no. 3, pp. 1445–1461, 2022

2022
[24]

Deep reinforcement learning-based edge caching in wireless networks,

C. Zhong et al., “Deep reinforcement learning-based edge caching in wireless networks,”IEEE Trans. on Cognitive Communications and Networking, vol. 6, no. 1, pp. 48–61, 2020

2020
[25]

Cooperative caching algorithm for mobile edge net- works based on multi-agent meta reinforcement learning,

Z. Wei et al., “Cooperative caching algorithm for mobile edge net- works based on multi-agent meta reinforcement learning,”Computer Networks, vol. 242, p. 110 247, 2024

2024
[26]

Cache-it: A distributed architecture for proactive edge caching in heterogeneous iot scenarios,

I. Zyrianoff et al., “Cache-it: A distributed architecture for proactive edge caching in heterogeneous iot scenarios,”Ad Hoc Networks, vol. 156, p. 103 413, 2024

2024
[27]

How to cache important contents for multi-modal service in dynamic networks: A drl-based caching scheme,

Z. Zhang et al., “How to cache important contents for multi-modal service in dynamic networks: A drl-based caching scheme,”IEEE Trans. on Multimedia, 2024

2024
[28]

Content popularity prediction towards location-aware mobile edge caching,

P. Yang et al., “Content popularity prediction towards location-aware mobile edge caching,”IEEE Trans. on Multimedia, vol. 21, no. 4, pp. 915–929, 2018

2018
[29]

Computing resource allocation in three-tier iot fog networks: A joint optimization approach combining stackelberg game and matching,

H. Zhang et al., “Computing resource allocation in three-tier iot fog networks: A joint optimization approach combining stackelberg game and matching,”IEEE Internet of Things Journal, vol. 4, no. 5, pp. 1204–1215, 2017

2017
[30]

Incentive mechanism for computation offloading using edge computing: A stackelberg game approach,

Y . Liu et al., “Incentive mechanism for computation offloading using edge computing: A stackelberg game approach,”Computer Networks, vol. 129, pp. 399–409, 2017

2017
[31]

Secure content delivery with edge nodes to save caching resources for mobile users in green cities,

Q. Xu et al., “Secure content delivery with edge nodes to save caching resources for mobile users in green cities,”IEEE Trans. on Industrial Informatics, vol. 14, no. 6, pp. 2550–2559, 2017

2017
[32]

Secure budget-aware edge caching in mobile social networks: A dynamic optimization approach,

Z. Seyedi et al., “Secure budget-aware edge caching in mobile social networks: A dynamic optimization approach,”Available at SSRN 5875412, APPENDIX Summary of Notations Notation Description CP Content provider ED Edge caching device MU Mobile user SCQ Secure Caching Quality ISet ofIEDs edi ED numberi, fori∈ {1, . . . , I} OSet ofOCPs cpo CP numbero, foro∈...