Pricing-Driven Resource Allocation in the Computing Continuum

Abstract: Deploying applications across the computing continuum requires selecting infrastructure nodes from geographically distributed and heterogeneous environments while satisfying constraints (e.g., performance, location). This decision problem is an important facet of resource allocation. As infrastructures grow in scale and heterogeneity, the resulting decision space becomes inherently combinatorial. Existing approaches typically formulate this problem as a constrained optimization task using ad-hoc representations of infrastructure topologies and demand, which hinders generalization across solutions. In contrast, Software as a Service ecosystems address a structurally similar configuration problem through pricings -structures whose plans and add-ons implicitly define the configuration space of possible subscriptions. Building on this observation, this work explores the potential of pricings as general-purpose representations of configuration spaces, positioning them as a promising alternative for addressing configuration problems, such as resource allocation, across the computing continuum. To this end, the paper presents the following contributions: i) a pricing-based formulation of the resource allocation problem in the computing continuum, enabling infrastructure configuration spaces to be represented using pricings; ii) a workflow that leverages PRIME, a pricing analysis engine, to explore these spaces and compute cost-optimal deployments satisfying functional and non-functional constraints; iii) generation processes for synthetic infrastructure topologies and workload demands; and iv) a dataset comprising 9,600 precomputed resource allocation scenarios to support benchmarking.

• The paper introduces a pricing-driven framework leveraging SaaS pricing abstractions to optimize resource allocation in heterogeneous, multi-provider environments.
• It employs a three-stage workflow—offer mapping, demand encoding, and optimization using the PRIME engine—to achieve cost-optimal deployments in sub-3 second median time for up to 200 nodes.
• Empirical evaluations on synthetic multi-provider scenarios demonstrate robust expressiveness, scalability, and cost-effectiveness for infrastructure planning across edge, fog, and cloud.

Pricing-Driven Resource Allocation in the Computing Continuum

Introduction and Motivation

Efficient resource allocation in the computing continuum—spanning edge, fog, and cloud—has become an inherently combinatorial and complex decision problem due to geographic distribution, infrastructure heterogeneity, and the need to satisfy both technical and business constraints. Existing approaches often rely on ad-hoc formulations for topology and demand representations, impeding solution generalization and comparison. This paper introduces a novel method leveraging the concept of "pricings"—standardized structures historically employed in SaaS ecosystems to represent configurable service offerings—as a general-purpose abstraction for modeling and optimizing resource allocation in heterogeneous, multi-provider environments (2604.12642).

Conceptual Framework

The authors frame the allocation challenge as the search for a cost-optimal configuration of nodes, mapped from infrastructure, capable of meeting aggregate demand under technical, economic, and governance constraints.

Problem Formulation

• Offer ($\mathcal{O}$): Infrastructure topology consisting of nodes (with resource vectors and contextual metadata) and a set of business rules (e.g., mutual exclusions, interoperability).
• Demand ($\mathcal{D}$): Aggregate resource vector required to service user workload for a region/conceptual deployment zone.
• Request ($\mathcal{R}$): Additional constraints such as provider whitelist, bounding node number, spatial limits, and price ceiling.

The objective is to identify a subset of nodes $C^\ast \subseteq \mathcal{N}$ that jointly satisfy all constraints while globally minimizing deployment cost.

Figure 1: An example of a typical pricing structure (here, Zoom) comprising features, usage limits, multiple plans, and add-ons.

Pricing-Driven Workflow

The principal contribution is a structured, three-stage workflow that employs machine-oriented pricing artifacts (iPricings) and an automated analysis engine (PRIME) to solve the allocation problem.

1. Offer Mapping

Infrastructure nodes and relationships are mapped onto a pricing domain:

• Each node → Pricing add-on
• Node types → Features
• Resource capacities → Usage limits
• Provider/business exclusions → Add-on exclusions

This abstraction encodes both topological flexibility and commercial constraints.

Figure 3: Projection of a sample infrastructure node into the pricing domain, demonstrating mapping from node capacities and business logic to add-on attributes.

2. Demand and Request Encoding

Demand vectors specialize usage limit constraints for each add-on, and concrete price expressions are instantiated per demand scenario—resource unit prices mapped to actual requirements. Requests introduce additional filters bounding the solution space (e.g., node budget, location).

3. Optimization and Back-Projection

The derived iPricing instance and scenario-specific filters are processed by PRIME, which resolves the CSP for cost-optimality; solutions are then projected back onto the original infrastructure.

Figure 2: End-to-end workflow illustrating infrastructure mapping to pricing, demand/request encoding, and optimization loop.

Experimental Evaluation

Comprehensive experiments are conducted using synthetic multi-provider topologies generated from the EUA dataset, focusing on both modeling expressiveness and computational viability.

Dataset and Scenario Synthesis

• Geographic scope: Primarily Melbourne region, with provider attribution (Telstra, Optus, Vodafone).
• Node resource profiles: Tiered (edge/fog/cloud), assigned plausibly to mirror real-world heterogeneity.
• Demands: Synthesized as aggregate vectors using stochastic models for active user rates, request throughput, shared/private memory, compute/accelerator needs, and storage.

  Figure 4: Geographical visualization of synthetic topology—each point represents a candidate infrastructure node.

Benchmarks and Parameters

The methodology is validated over 9,600 distinct deployment scenarios, spanning four application types (CCTV, VR, robotics, LiDAR) and systematically varying demand intensity and topology scale. For each scenario, 100 diverse topologies are sampled to ensure robust statistics.

Results

• Modeling expressiveness: All scenarios—including multi-provider, multi-mode, and interoperability constraints—are expressible via the proposed abstraction without ad-hoc workarounds.
• Optimality and feasibility: PRIME consistently finds feasible, globally optimal deployments or correctly identifies infeasible scenarios (e.g., over-constrained requests).
• Scalability: Execution time remains stable with increasing demand. Computation time scales with candidate node count; for up to 200 nodes, median solving time remains sub-3 seconds—suitable for interactive or near-real-time planning.

Figure 5: Execution time of the workflow under increasing problem size; median values across multiple topologies show bounded growth and negligible variance up to practical infrastructure sizes.

Theoretical and Practical Implications

This work establishes evidence that iPricing-based abstractions are not limited to business-layer SaaS configuration but can generalize to infrastructure resource allocation across the continuum. The ability to represent both technical and business/gov constraints compactly enables transfer and reuse of existing analysis tools—bridging a modeling gap between service-level variability and infrastructure-level compositionality.

On the practical side, the workflow allows automated, cost-aware, multi-provider resource planning, simplifying integration of pricing logic and technical constraints. Execution times suggest direct applicability to operational decision support for infrastructure deployments.

Limitations and Future Work

The experiments rely on synthetic resources and business constraints due to the lack of perfectly matched real-world datasets. Evaluation focuses on a single metro-scale region and a triad of providers, potentially limiting external validity. The proxies employed (e.g., geo-distance as a latency surrogate) simplify certain network realities.

Future research should expand to real-world provider data, broader geographic scales, and production workload traces. Incorporating online monitoring and re-optimization, generalizing the abstraction to other distributed computing domains (e.g., federated learning, energy-sensitive scheduling), and developing hybrid/distributed CSP solving approaches are promising directions for increased robustness and scalability.

Conclusion

The paper demonstrates that pricing-driven abstractions, originally confined to SaaS business logic, directly enable expressive and efficient resource allocation in the computing continuum. The proposed workflow, underpinned by iPricings and automated CSP analysis, shows empirical tractability and flexibility in multi-provider, resource-diverse environments. This modeling unification invites further exploration of cross-layer abstractions and tool reuse across the cloud-edge continuum, with implications for both research and operational infrastructure management.