arXiv CSopen access

Urban Limits as Design Constraints: Identifying Suitable Locations for Distributed, Photovoltaic-Powered Servers

Urban Limits as Design Constraints: Identifying Suitable Locations for Distributed, Photovoltaic-Powered Servers Justin Chikhaoui

[email protected] LIRMM, Univ. Montpellier - CNRS Montpellier, France

arXiv:2606.18940v1 [cs.DC] 17 Jun 2026

Daniel Siret

[email protected] Nantes Université, ENSA Nantes, École Centrale Nantes, CNRS, AAU-CRENAU, UMR 1563 Nantes, France

Abstract Urban territories face growing tensions between increasing digital demand, limited resources, and socially constrained built environments. Although distributed computing paradigms such as edge and fog computing are widely presented as solutions for reducing latency and energy dissipation, the scientific literature largely overlooks where such infrastructures can be physically and socially deployed within cities, and typically neglects urban constraints, environmental impacts, and equity considerations. This paper proposes a methodology for identifying suitable urban locations for deploying distributed servers under structural, environmental, and social limits. Relying exclusively on existing infrastructures and anthropised surfaces, it combines legal frameworks, ongoing urban projects, citizen consultations, and scientific literature to construct a place-based glossary of viable site typologies, evaluated through energy, spatial, and qualitative criteria. Applied to the French city of Montpellier, our results illustrate how urban constraints and local resources shape the feasibility of decentralised, solar-powered digital infrastructures, and highlight the value of territorialised approaches for rethinking digital services within urban limits.

Keywords Urban digital infrastructure, digital sobriety, territorial methodology, edge and fog computing, photovoltaic energy harvesting

1

Introduction

Urban territories constitute a fundamentally complex framework, characterised by a dense and heterogeneous built environment, a plurality of actors, and a living space shaped by sometimes contradictory requirements [91]. Within this space, the production of the city consists of responding to collective needs — education, health, culture, mobility, economic activities — while navigating material, social, and political constraints [18]. Increasingly, the response to these needs is mediated by digital services: sensors, applications, platforms, control Workshop on Computing within Limits, June 23-25, 2026, 2026.

Thomas Leduc

[email protected] Nantes Université, ENSA Nantes, École Centrale Nantes, CNRS, AAU-CRENAU, UMR 1563 Nantes, France

Abdoulaye Gamatié

[email protected] LIRMM, Univ. Montpellier - CNRS Montpellier, France systems, and decision-support tools. This digitisation dynamic sustains the idea that technology could respond ever more precisely to human needs [73], promising an enhanced understanding of urban systems and their actors, e.g., through land-use strategies that integrate a holistic view of the territory. Recent decades have, however, seen the rise of a technosolutionist vision embodied by initiatives such as smart cities [105], which promote the massive extension of sensors and digital services as a privileged approach to solving urban and societal problems. This vision is not neutral: it entails significant environmental and social impacts [107, 108]. Among the most widely deployed digital infrastructures, data centers require substantial material inputs during both construction and operation. The extraction, transformation, and transportation of these materials exert growing pressure on natural reserves and the territories where they are located [112]. On the social dimension, the widespread deployment of digital devices reshapes daily practices, social relations, and conditions of access to urban services, fostering dependency on technical infrastructures and reinforcing the vulnerability of populations with limited digital access or skills [83]. These observations raise a central question: how can digital needs — not all of which are necessarily desirable — be met without exceeding the environmental and social limits of urban territories, limits that existing material infrastructures already displace and exacerbate? Our vision. Several strategies address the tension between meeting digital needs and managing their associated impacts [81, 87]: sobriety, degrowth, the abandonment of certain services, or the redefinition of needs, among others. Here, we focus on an approach that leverages specific characteristics of the urban environment. We identify at least two properties that enable such an objective: (1) the judicious reuse of the already urbanised and artificialised built environment; and (2) the exploitation of urban solar resources. By “judicious reuse”, we refer to a deployment model in which digital demand and the computational resources that serve it are co-located, while remaining within the bounds imposed by the integration of infrastructure into the built urban environment (notably its institutional, architectural, and social constraints). Two further elements are considered in order to meet digital needs without deteriorating other dimensions of the territory. The

Workshop on Computing within Limits, June 23-25, 2026, ,

first concerns the spatial limits of cities: our approach favours locally grounded responses to local needs, anchoring our work within the specific challenges of a fine-grained territorial scale through a bottom-up strategy [99]. The second concerns the structural specificities of urban territories — building ownership, infrastructure functions, social acceptability, and so on — which call for a deployment approach that is sensitive to the living and working conditions of the territory, ensuring that computing units are integrated with due regard for local actors.In practice, this translates into the deployment of computing units powered by photovoltaic energy on existing infrastructures and buildings. Reusing existing buildings prevents further land artificialisation and the associated environmental impacts, while solar energy minimizes dependence on fossil fuels.This strategy may nonetheless generate unintended consequences [77].The first is the potential proliferation of digital infrastructures and services. The second is the displacement of the primary social functions of the selected sites: urban furniture and public facilities serve as spaces of rest and collective appropriation that may be compromised by the technical requirements of hosting computing units, particularly in dense areas where public space is already limited. technological expansion effect, we propose deploying computing units only in situations of infrastructure renewal, thereby avoiding the creation of new infrastructures that do not respond to a specific need — a situation that could otherwise stimulate digital demand and reinforce the associated environmental and social impacts [3, 68]. Furthermore, the intermittency of solar energy restricts continuous operation and increases maintenance needs, thereby reducing incentives for intensive exploitation and encouraging more restrained patterns of use. Together, these constraints lead us to select urban deployment locations with careful attention to the existing uses of infrastructures and buildings. Contribution and outline. This paper makes a twofold contribution. The first is a methodology aligned with the vision presented above, based on the identification and characterisation of suitable locations in urban environments so as to better account for the limits and dynamics of urban spaces. The second is the application of this methodology through a case study focused on the city of Montpellier, in southern France. The paper is organised as follows. We first review the relevant literature (Section 2), then detail the proposed methodology (Section 3), before presenting a case study that allows us to identify a typology of urban locations suited to the digital needs specific to Montpellier based on our methodology (Section 4). Finally, we give concluding remarks and discuss some future research directions and potential applications (Section 6).

2

Placement of computing units: a problem mainly framed as algorithmic optimization

The scientific literature offers a substantial body of work seeking to address the tension between digital demand and its territorial impacts [35, 41, 67, 79] through the deployment of distributed architectures. In particular, edge and fog computing are presented as paradigms that bring computing capabilities closer to points of use, thereby reducing latency, network congestion, and the energetic distance between data production and processing [110].

Chikhaoui et al.

Distribution of computing units: a strategy aimed at promoting sustainability. Distributed infrastructures, such as edge computing systems, able to address the bandwidth and latency issues of centralized infrastructures, in particular by reducing congestion on long-haul networks such as submarine fibre-optic cables. They are often described as inherently more sustainable than the centralized ones, but this claim is contested. Several studies show that sustainability outcomes depend less on the spatial proximity of computing resources than on how infrastructure deployment, capacity provisioning, and the surrounding energy system are configured [7, 15, 58, 100]. The environmental benefits of distribution are therefore not intrinsic; they depend on these systemic choices. The literature points to two main levers for improving the sustainability of edge and fog computing [110]: powering distributed infrastructures with renewable sources [29], and co-locating computing units with energy production sites to limit transmission losses [82]. The distributed nature of edge and fog computing also makes it possible to exploit territorial heterogeneity, such as locally available renewables or favourable climatic conditions, and to rethink infrastructure design beyond network-centric criteria. Passive cooling through outdoor immersion of edge nodes has been evaluated [48], and other work proposes reusing the waste heat of distributed units to supply urban thermal demand [33]. Distribution of computing units: a strategy aimed at promoting sustainable development. Distributed computing was initially conceived to address the bandwidth and latency limitations of centralized infrastructures, notably by reducing congestion in communication networks such as submarine fiber-optic systems. Although distributed paradigms are often portrayed as inherently more sustainable than centralized computing, this assumption remains controversial. Several studies indicate that sustainability outcomes depend less on the spatial proximity of computing resources than on the interplay between infrastructure deployment, capacity provisioning, and the characteristics of the associated energy systems [7, 15, 58, 100]. In this sense, the environmental benefits of distribution cannot be considered intrinsic, but rather contingent upon broader systemic configurations. In parallel, the literature identifies several strategies that could enhance the sustainability potential of edge and fog computing. As emphasized in the literature [110], energy-efficient approaches are essential to mitigate the environmental impact of these distributed paradigms. Two main directions can be identified: on the one hand, the integration of renewable energy sources to power edge and fog infrastructures [29], and on the other hand, the co-location of computing units with energy production sites, which may reduce transmission losses and improve overall energy efficiency [82]. More broadly, the distributed nature of edge and fog computing offers the opportunity to leverage territorial heterogeneity, for instance by exploiting locally available renewable resources or favorable environmental conditions. This perspective enables a rethinking of infrastructure design beyond purely network-centric considerations. For example, passive cooling strategies based on the outdoor immersion of edge nodes have been explored [48], while other works propose reusing the waste heat generated by distributed computing units to meet urban thermal demands [33].

Urban Limits as Design Constraints: Identifying Suitable Locations for Distributed, Photovoltaic-Powered Servers

Geographical distribution of computing units: a central yet under-explored issue. Surveys on distributed infrastructures consistently show that the literature has concentrated on architectures, resource management, security, and orchestration [1, 52, 75], leaving a significant blind spot: the geographical positioning of computing units. Yousefpour et al. [110] explicitly note that, despite its importance for quality of service and infrastructure sustainability, this topic remains under-studied due to the absence of standardised frameworks and mature protocols. The works that do address geographical placement generally model it as a multi-objective optimization problem, integrating criteria such as latency, load balancing, cost, or energy consumption [20, 65]. These approaches, however, share recurring limitations. They typically rely on pre-defined locations — base stations, roadside units, or access points — which restrict the range of sites considered [8]. They also tend to adopt incomplete territorial models, which may yield theoretically optimal placements that are urbanistically infeasible [13]. Digital demand is moreover often estimated indirectly, either through generic network communication flows [104] or population-based territorial models [8]. Taken together, these tendencies reveal a persistent dissociation between computational optimization and territorial embeddedness — a dissociation that stands in tension with the necessity, recalled in the introduction, of articulating digital services with the material and social realities of the urban environment. Integrating the urban territory into the placement problem. A number of recent studies have nonetheless begun to engage more seriously with the urban territory. Several highlight that existing infrastructures can constitute deployment opportunities for computing units [5, 40, 61, 70], offering advantages such as structural or energy accessibility, proximity to usage contexts, pre-existing electrical supplies or communication devices, and, in the case of rooftops, favourable solar exposure. Beyond static infrastructure, other contributions incorporate spatial and temporal dynamics [22, 24, 90]. Population densities, variations in digital demand, and daily mobility patterns all affect where computing capacity is most relevant, and accounting for these dynamics helps avoid rigid deployments ill-suited to territories characterised by heterogeneous human flows and uneven digital practices. A further strand of work underlines the importance of urban constraints — physical obstacles, architectural properties, and failure risks — which can affect signal propagation and network robustness [88, 89]. These studies make clear that the reliability of a distributed computing network is not solely an algorithmic property but also a material and spatial one. Finally, certain contributions, such as [40], explicitly seek to articulate deployment with urban renewable energy, modelling the solar potential of rooftops to size locally powered computing nodes and demonstrating that a convergence between digital infrastructure and local energy resources is achievable. Limitations of existing work. Despite these contributions, research on the urban placement of computing units remains fragmented across disciplines and approaches. Each study tends to isolate a single territorial dimension without offering a framework that encompasses the multiple material, social, and energetic layers

Workshop on Computing within Limits, June 23-25, 2026, ,

conditioning deployment feasibility. No existing work aims to construct a broad, systematic, and characterised inventory of urban locations capable of hosting computing units. Furthermore, the reliance on indirectly estimated digital demand introduces biases with potentially significant consequences. Oversizing decentralized infrastructures may encourage increased digital usage and triggering a technological pull effect. Conversely, undersizing them may fail to meet actual urban needs and may weaken essential territorial functions. Both outcomes can deepen inequalities among inhabitants, as some would see their needs adequately addressed while others remain marginalised from equitable access to digital services. More broadly, the literature reveals three persistent gaps: the geographical placement of distributed computing units remains under-explored outside information and communication sciences; the specific characteristics of urban territories represent major yet still under-exploited levers; and existing approaches account for only a fraction of the dimensions required to design infrastructure that is truly situated, sustainable, and compatible with urban limits. Our work positions itself precisely within this gap, by simultaneously accounting for potential urban locations, their structural and energetic properties, and the nature of the digital needs that the infrastructure must address. Beyond the technical literature on distributed computing, the LIMITS research community has already begun to address many of the questions raised in this work, particularly regarding how digital systems can be designed under environmental and social constraints. Prior work in computing within limits emphasizes the need to reconsider the growth-oriented paradigm of digital infrastructures, advocating instead for approaches grounded in sufficiency, resource awareness, and the acknowledgment of material limits [76, 92, 101]. In this perspective, notions such as digital sobriety, rebound effects, and the redefinition of needs have been explored as central challenges for sustainable computing [10]. Moreover, LIMITS research stresses the importance of situated and place-based approaches, where infrastructures are designed in relation to local contexts, resources, and communities rather than as abstract, globally optimized systems [51]. Our work builds on these conceptual foundations by operationalizing them in the context of urban digital infrastructure siting, proposing a methodology that explicitly integrates spatial, social, and energy constraints into the identification of deployment locations. In doing so, it contributes to bridging the gap between the critical, conceptual perspective developed within LIMITS and the more operational questions of infrastructure design and placement. The LIMITS community has already addressed several of the questions this work raises, particularly how digital systems can be designed under environmental and social constraints. A prior work argues for moving away from the growth-oriented paradigm of digital infrastructures and toward approaches grounded in sufficiency, resource awareness, and material limits [76, 92, 101]. Within this stream, digital sobriety, rebound effects, and the redefinition of needs have been examined as central challenges for sustainable computing [10]. LIMITS research also calls for situated, place-based approaches, in which infrastructures are designed w.r.t. local contexts, resources, and communities rather than as abstract globally optimized systems [51].

Workshop on Computing within Limits, June 23-25, 2026, ,

We build on these foundations and operationalize them for urban digital infrastructure siting, proposing a methodology that integrates spatial, social, and energy constraints into the identification of deployment locations. The contribution thus connects the critical perspective developed within LIMITS to the operational questions of infrastructure design and placement.

3

Methodology for computing unit deployment

Our methodology aims to identify suitable locations for deploying distributed computing units in urban environments, motivated by the absence of a rigorous framework in the scientific literature for determining which sites are appropriate for such infrastructures. It guides location selection toward buildings and surfaces that respect the structural and environmental limits of the urban territory through two explicit constraints. The first is a material constraint: reducing the environmental footprint of digital technologies by relying strictly on existing buildings and infrastructures. The second is a social constraint: avoiding conflicts of use by selecting only locations whose technical or multipurpose function can accommodate a digital service without disrupting their primary role.

Figure 1: Methodology for identifying relevant locations for the deployment of distributed server systems within the structural limits of the urban territory. The methodology unfolds in two steps (Figure 1). The first identifies candidate locations that satisfy urban acceptability constraints, producing a glossary of viable site typologies. The second characterises these locations in order to select those best suited to digital, energy, and social needs. These steps must be understood within their territorial context: grounding the search in a specific territory makes it more realistic by incorporating local specificities. Legislation governing rooftop usage, for instance, differs between different countries. Similarly, the physical morphology of computing units conditions which locations are feasible: a small IoT-type device may be installed at a roadside traffic signal, whereas a server equipped with memory, storage, and a GPU requires a larger dedicated space.

Chikhaoui et al.

The glossary produced by the first step constitutes a foundational element of our approach, directly addressing a gap in the literature on location types suitable for distributed server deployment. It takes the form of a dataset of locations characterised by their typology and accompanied, at minimum, by their geographical coordinates. The second step takes as input this glossary together with a digital demand defined both spatially and quantitatively. Proper qualification of this demand is essential to ensure proportionality: an excessive deployment of servers risks inducing a rebound in digital consumption [3], whereas insufficient deployment would fail to meet actual needs. Through this methodology, we do not seek to optimise the deployment of distributed servers according to classical metrics — latency, performance, or load balancing. Rather, we aim to assess the actual capacity of the urban territory to host a form of digital technology that is more environmentally sober and socially acceptable.

3.1

Identification of suitable locations

The first step takes as input a specific territory features and a computing unit morphology (e.g., servers) to be deployed; several morphologies may also be considered simultaneously. Identification. Because the search for locations is bounded by both a territory and a server morphology, we focus on sites that may satisfy these two constraints. The methodology therefore draws on typologies of locations already in use within the territory for comparable deployments, identified through four sources: law and regulation, existing projects, citizen consultation reports, and scientific literature. Given that the deployment of distributed servers is rarely discussed explicitly in these sources, we adopt the following working hypothesis: servers may be installed in the same locations as solar panels, such that a panel may be accompanied by a computing unit without requiring additional space, as illustrated by Genesis systems [30, 86]. Genesis (short for "Generation of Energy-Synergistic Information Systems") (Figure 2) exemplifies a green computing paradigm. It is an electronic device that integrates at least one solar panel with energy transfer logic and can optionally include batteries and compute servers with corresponding data transfer capabilities. Its solar panels enable local energy harvesting, while the Genesis computing systems could also serve as a cloud-based Platform-as-a-Service (PaaS) infrastructure. The Genesis paradigm addresses the resource-intensity computing services by aligning compute with the conditions of its host environment rather than expanding dedicated infrastructure. It has three properties that support this alignment. First, placing computing units next to photovoltaic production reduces transmission losses and lets capacity track the local energy profile. Second, outdoor deployment with a suitable enclosure provides natural ventilation, which removes most of the active cooling components and the energy they consume. Third, mounting units on existing urban infrastructure rather than building dedicated facilities avoids further urban sprawl, a direct driver of biodiversity loss. Law and regulation inform the obligations, recommendations, and constraints applying to any intervention in urban development [14]. Obligations may identify areas constrained for deployment;

Urban Limits as Design Constraints: Identifying Suitable Locations for Distributed, Photovoltaic-Powered Servers

Workshop on Computing within Limits, June 23-25, 2026, ,

overlooking relevant site types: the broader the glossary, the greater the flexibility of deployment, which in turn supports a resilienceoriented approach toward urban limits [23]. All typologies identified through this process share the characteristics of being anthropised surfaces, largely non-conflictual, compatible with local photovoltaic supply, and integrable without significantly extending the digital material footprint.

Figure 2: Genesis prototype deployed at the University of Montpellier recommendations reveal locations where deployment would be supported by territorial governance; and constraints indicate sites where deployment would be prohibited or difficult. For example, Articles L111-19-1 and R111-25-17 of the French urban planning code identify parking areas as locations where renewable-energy devices are required [38, 39]. Urban projects conducted by local authorities, transport operators, companies, and territorial organisations provide concrete grounding for the reflection on digital materiality and the need to rely on resources already present within a territory [102]. Rather than adding dedicated infrastructures, urban experiments show that cities already contain numerous locations capable of hosting energy devices and, by extension, distributed computing units [32]. They confirm that the relevance of a solar-first deployment depends on identifying existing, stabilised, low-conflict surfaces. Among such examples, the Wattway project demonstrates the feasibility of integrating photovoltaic panels directly into the pavement of cycling lanes [26]. Citizen consultation reports, although rarely accessible, provide insight into how inhabitants appropriate and reshape the territory to match their expectations. They therefore represent a source of locations that have been discussed and validated by those most directly concerned, often accompanied by motivations that shed light on citizen expectations and reservations regarding the evolution of their living environment [85]. Finally, the scientific literature broadens the search through conceptual and applied research on server deployment. Although not always realistic, these contributions support innovation by articulating strategies around clearly defined objectives, often intersecting with technological challenges [40]. Locations such as lampposts [6] and bus stops [69] are identified in this way. Together, these four sources enable the construction of a coherent set of potential urban locations for distributed server deployment. Identification thus becomes an exploratory process in which candidate locations are progressively revealed through the analysis of diverse documentary sources. The typologies identified must, as noted above, be compatible with the territory of interest. Territorial adaptation is therefore possible and encouraged: if metro station rooftops are identified as potential locations [71], a city served only by tramways may consider tram stations instead. Such adaptation is essential to avoid

Selection. Not all identified locations automatically qualify for the glossary. A selection is required to discard typologies that are structurally incompatible with server deployment, based on four criteria: material compatibility, structural regulations, site specification, and site knowledge. The material criterion concerns incompatibilities between the server morphology and the physical characteristics of the site. These may arise from site types that compromise server operation, insufficient structural capacity to support a server, or inadequate space to accommodate computing units. The structural criterion reflects the strict regulations governing urban development [96]. A deployment may conflict with these rules: in France, for instance, listed historical buildings are protected and require authorisation from the Architectes des Bâtiments de France for any modification to their exterior appearance [42], making the installation of digital infrastructure on such buildings particularly difficult. The specification criterion addresses the risk of over-broad categories. Some location types are too heterogeneous to be expressed uniformly: scientific publications may propose deployments at network access points [62, 66], which can take diverse forms — antennas, routers, and so on — making it impossible to assess material compatibility across the category as a whole. Finally, the knowledge criterion concerns data availability. Selecting appropriate locations requires a sufficiently comprehensive view of the territory [21]; without reliable data describing the geographic position of sites, it is only possible to work on a case-by-case basis, precluding any territorial-scale assessment.

3.2

Characterisation of identified locations

The second step serves as a decision-support tool for identifying, within the glossary, the most favourable locations for server deployment. Rather than optimising classical computational metrics — latency, load balancing, and so on — it aims to determine which locations: (1) can meet the defined digital demand; (2) provide a local response to a local need, thereby remaining within context without drawing on external resources; and (3) enable server deployment while leveraging the advantages of each typology and acknowledging its constraints. Three criteria structure this characterisation: energy, spatial, and qualitative. Energy criterion. The energy criterion assesses each location typology’s capacity to generate sufficient photovoltaic energy to sustain a digital service, in keeping with the objective of minimising environmental impact. A photovoltaic (PV) panel converts shortwave solar irradiance into electricity through semiconductor cells: when photons strike these cells, they excite electrons and generate an electric current [9]. Panel efficiency depends primarily on cell

Workshop on Computing within Limits, June 23-25, 2026, ,

Chikhaoui et al.

quality, which governs the conversion ratio between received irradiance and produced electricity — typically ranging from 12% to 17.5% depending on the technology selected [80] — and on the solar irradiance received, which is itself affected by geographic position, cloud cover, and shading [84]. Geographic position determines the angle between the sun and the receiver as well as the distance separating them, both of which condition the received irradiance [2]. Combined with established models of the sun’s apparent trajectory [47, 53, 63], the location of the studied sites allows for estimating the cumulative irradiance over a given period of time (i.e., the irradiation named irradiationcs ) under clear-sky conditions. Applying a conversion rate of 15% yields a theoretical energy production estimate under optimal conditions (see Eq. 1): prod = 0.15 × irradiationcs

(1)

Under real conditions, cloud cover reduces the fraction of irradiance reaching PV panels. We therefore introduce in Eq. 2 a variable 𝑛𝑒𝑏 ∈ [0, 1] representing this fraction: prod = 0.15 × irradiationcs × neb

(2)

Urban environments are further characterised by heterogeneous building heights and forms that generate variable shading conditions. To account for this, we introduce in Eq. 3 a binary shading variable 𝑠ℎ, set to 0 if the site is shaded during the considered hourly period and 1 otherwise: prod = 0.15 × irradiationcs × neb × sh

(3)

Accounting for these three factors enables the estimation of PV energy production for each location typology. Given a digital demand and its associated energy consumption, it then becomes possible to right-size a distributed server network across the territory, creating a response environment for digital needs that minimises the environmental impact of computing energy consumption. Spatial criterion. The spatial distribution of potential locations conditions how the infrastructure articulates with the city. We define a coherent spatial distribution through the principle of locality between digital demand and the infrastructure meeting it — expressed as the capacity to meet one’s needs without infringing on the capacity of one’s neighbours to meet theirs [19]. In practice, this translates into analysing the distance between proposed locations and digital demand: by minimising these distances, we seek to foster spatial equity [97], limit energy losses associated with data and energy transport [16], and reduce latency. This framing also opens a discussion on forms of digital sufficiency that may emerge when needs cannot be satisfied locally [57]. Distance analysis is operationalised through two complementary indicators. The first estimates the maximum distance required for one or more location typologies to meet a given digital demand, enabling comparison across typologies and prioritisation of those requiring the smallest travel distances. The second examines how the capacity to satisfy demand varies as a function of distance, revealing the irregular spatial availability of locations across the territory. A typology may require a large maximum distance to meet total demand while still satisfying a substantial fraction of it at short range — a finding that can support a redefinition of the initial digital demand toward lower levels that can be met as locally as possible,

thereby anchoring digital sobriety within a spatial reflection on urban limits. Qualitative criterion. The urban territory cannot be reduced to quantitative criteria alone [37]. Our methodology therefore incorporates qualitative aspects that capture the intrinsic properties of location typologies in the context of server deployment, along two axes of common qualification and a set of typology-specific properties. The first axis is environmental, focusing on the structural modifications required to deploy distributed servers: the works needed to render a site compatible and the need to connect existing buildings or infrastructure to communication and electrical networks. In the absence of such connections, additional environmental and economic costs are incurred [94]. Connection to the main electrical grid also offers a twofold advantage: it enables the valorisation of surplus generation through resale and ensures continuity of supply during periods of solar intermittency. The second axis concerns deployment acceptability, captured through the number and nature of owners involved [50]. Engaging multiple stakeholders is inherently more complex than dealing with a single one, whether at the scale of a single co-owned building requiring unanimous resident approval or across multiple locations divided between public and private actors. Furthermore, the conditions differ substantially depending on the nature of ownership: a French municipality may perceive collective or environmental value in hosting computing units, whereas a private individual may be deterred by costs, lack of expertise, or concern over impacts on their property. These two axes yield six criteria across which typologies are scored. Assuming equal importance across criteria, each (criterion, typology) pair receives a score from the discrete set {0; 0.5; 1}, where 1 indicates strong performance, 0.5 average performance, and 0 weak performance. This scoring permits a hierarchical ranking of typologies. Beyond these common properties, each typology may also present specificities that are crucial to consider but do not permit fair comparison across categories. This limitation stems from the inherently unique nature of urban situations: a fully systemic treatment of the urban territory is not possible, and a situated consideration of the territory and its actors remains necessary to assess the relevance of the locations selected.

4

Case study

In the next, we first describe the experimental setup of the case study. Then, we show the typical results provided by our methodology.

4.1

Experimental setup

We illustrate our methodology through a case study conducted in Montpellier (INSEE code 34172), a city in southern France covering 56.88 km2 with a population of 516,657 inhabitants. Digital demand. The digital demand studied here is the use of large language models (LLMs) in secondary education. Chatbots have been recently adopted by high-school students for learning,

Urban Limits as Design Constraints: Identifying Suitable Locations for Distributed, Photovoltaic-Powered Servers

writing, information retrieval, and similar academic tasks [95]. Unstructured and unrestricted use of these technologies can be harmful: it has been linked to lower academic performance, reduced neural connectivity patterns, and memory erosion. When their use is framed by a strict pedagogical protocol, the opposite effects are observed, with gains in learning and motivation [36, 59, 60]. Three conditions make this difference: the model is used as a guide rather than a source of direct answers, it is consulted only after an initial phase of independent thinking, and it is restricted to low-level support such as spell-checking. Under close supervision by teaching staff, classroom use of chatbots can therefore deliver benefits. Deploying an LLM by and for high schools is well suited to this framing. LLM inference is, however, energy-intensive: the electricity consumption of the infrastructure needed to meet AI demand is projected to grow by a factor of two to fourteen between 2025 and 2035 [78]. At that scale, the infrastructure supporting conversational agents risks conflicting with current environmental goals such as limiting global warming. LLMs can thus play an active role in school education, but only under conditions. Their deployment cannot be planned on the assumption of unlimited resources. Reconciling the social benefits of these tools with global constraints requires an infrastructure that supports their use within environmental limits, and a usage regime designed to limit rebound effects. The demand is centered on the seven public high schools of Montpellier (Figure 3): Joffre, George Clemenceau, Jules Guesde, Jean Mermoz, Françoise Combes, Agropolis, and George Frêche. We define a scenario in which 10% of students use an LLM between 08:00 and 18:00 (UTC+1) on 21 December (winter solstice). According to the French Ministry of Education, this corresponds to 158, 115.2, 197.2, 177.4, 25.2, 39.1, and 12.8 active users per school, respectively.

Figure 3: City of Montpellier and its public high schools. Computing infrastructure. We adopt a computing unit morphology based on the Genesis paradigm [40]. This approach allows digital infrastructure to take various forms so as to adapt to different location typologies while being equipped with a solar panel and able to operate outdoors. To quantify the energy required to meet the digital demand of the high schools, we use a Genesis server prototype deployed in Montpellier, equipped with an Intel Xeon E3 3.3 GHz processor and 16 GB of RAM, without a GPU or TPU. Based on a consumption study, we determine an average power draw of 502 Wh when running the Llama 3.1 8B model. All processing is carried out in a Python 3.10.12 environment.

Workshop on Computing within Limits, June 23-25, 2026, ,

Table 1: Glossary of identified and modeled location typologies. Location type

Count

Data source

Base station Bench Industrial building Service building Sports building Traffic light Industrial wasteland Streetlight Bike storage room Multi-family housing Single-family housing Parking area Polluted site Bus stop Tram stop Table

599 4,259 1,207 12,728 251 1,638 100 87,449 264 18,645 48,037 2,126 42 816 297 712

Mon Réseau Mobile 1 OSM 2 BD TOPO® 3 BD TOPO® BD TOPO® OSM CartoFriche 4 Open Data Montpellier 5 OSM BD TOPO® BD TOPO® OSM Sites et sols pollués 6 OSM OSM OSM

We chose not to include an energy-storage system in the target system design, so that servers operate only during daylight hours. Two reasons motivate this choice. First, urban energy storage relies mainly on batteries [4]. The extraction of the required resources, the manufacturing of the devices, and their transport carry heavy environmental costs – in particular for the movement of raw materials [74] – and heavy social costs, including the forced displacement of populations linked to certain mining activities [72]. Second, restricting access to daylight hours regulates usage and reduces the risk of addiction. Continuous energy availability would encourage intensive LLM use during periods normally reserved for student rest, which is necessary for the balance of daily rhythms [109]. Location typologies. We identify candidate hosting locations through French legislation and regulation [43–46], urban projects deploying photovoltaic panels on existing infrastructures [11, 12, 25, 34, 56, 64, 98, 103, 106, 111], and scientific studies [5, 40, 61, 62, 69]. Citizen consultation documents are omitted owing to limited data availability. After search and selection, sixteen location typologies are retained: lamppost, individual housing, collective housing, service building, bench, parking area, industrial building, bus station, traffic light, table, network antenna, bicycle facility, sports building, tram station, brownfield, and polluted land. For each typology, all instances within the administrative boundaries of Montpellier are modeled in Python with Geopandas [55], including their geographic position and, where applicable, their geometric footprint (Table 1). For typologies lacking geometric data, contextually coherent installation areas are assigned (Table 2). 1 Mon Réseau Mobile details the list of mobile network sites in France.

2 OpenStreetMap (OSM) data consists of collaboratively contributed geospatial features

describing a wide range of real-world geographic entities. 3 The BD TOPO® dataset is a database describing France’s infrastructure.

4 CartoFriche lists potentially unused or derelict artificialized sites across France.

5 Open Data Montpellier provides open access to a wide range of public datasets from

the city 6 Sites et sols pollués identifies polluted or potentially polluted sites in France.

Workshop on Computing within Limits, June 23-25, 2026, ,

Chikhaoui et al.

Table 2: Proposed installation area for location typologies without geometric data. Location typology

Installation area (m2 )

Base station Bench Table Streetlight Bus stop Tram stop Bike storage room

3 2 2 1 4 4 4

Solar irradiance and shading. Clear-sky irradiance is estimated at hourly intervals using the PVLIB library [54] and the Ineichen model [53]. Shading is computed from a digital surface model provided by the French national geographic institute, following the methodology of Corripio [28] with insolation a Python library [27]; each location is assigned a binary shading variable (0 if shaded, 1 otherwise) at each hour. To account for cloud cover, data from Météo France describing the number of sunny hours per month are extrapolated to an intra-day distribution, concentrating cloud-free periods at peak irradiance hours (11:00–15:00 UTC+1).

4.2

Results

Energy feasibility. Under the meteorological scenario described above, eight location typologies are able to sustain the energy demand associated with the LLM (Figure 4). Each of these typologies produces sufficient energy during the five hours of peak sunlight. 5

number of hours

4 3 2

0

Base station Single-family housing Multi-family housing Service building Industrial buildings Sport building Traffic light Parking Bench Table Streetlight Bus stop tram stop Bike storage room Industrial wasteland Polluted site

1

Figure 4: Number of hours for which each location typology can sustain the LLM energy demand.

Proximity to demand. Among the energy-feasible typologies, we assess the ability of each to meet the digital demand locally. A linear solver minimizes the distance between candidate locations

and high schools while ensuring the energy required for operating the computing units is satisfied without exceeding the site’s capacity. Figure 5 presents, for each of the five peak solar hours (11:00– 15:00 UTC+1), the cumulative digital demand coverage as a function of the maximum distance allowed between a computing unit and a high school. Each curve corresponds to one location typology; the filled circle at its right extremity marks the minimum distance at which that typology alone achieves full demand coverage. Reading a curve from left to right therefore reveals how quickly — in spatial terms — a typology can respond to the totality of the digital demand: a steep rise close to the origin indicates that a large share of demand can be met at very short range, while a gradual slope or a late saturation point signals a typology whose suitable instances are sparsely distributed relative to the demand sites. The replication of this representation across five hourly panels reflects the temporal variability of photovoltaic production: the number of computing units that can actually be powered at a given location changes as solar irradiance evolves throughout the day, which in turn affects how much demand each typology can cover at a given distance. The overall spatial ranking among typologies nevertheless remains stable across all five hours, attesting to the robustness of the conclusions drawn below. Two groups of typologies emerge clearly. The first comprises service buildings, single-family housing, multi-family housing, and parking areas, all of which reach full coverage within less than 500 m across every hourly panel. Service buildings saturate at the shortest distances — between 35 and 85 m depending on the hour — because the rooftops of the high schools themselves belong to this typology, making the spatial gap between demand and supply negligible. The remaining three typologies in this group achieve full coverage between 274 and 445 m, indicating that sufficiently dense instances of each exist in the immediate urban fabric surrounding the demand sites. The second group — industrial buildings, sport buildings, streetlights, and polluted sites — requires distances exceeding 1 km to satisfy total demand, with polluted sites reaching saturation only beyond 4.5 km. Given the availability of closer alternatives, these typologies are excluded in order to preserve the local response strategy central to our methodology. Qualitative assessment. The four selected typologies are then evaluated qualitatively through a hierarchical scoring based on the need to connect locations to power and internet networks, required structural modifications, number of distinct owners, and nature of ownership (Table 3). This evaluation places service buildings in first position. In the specific context of public high schools, deploying computing units on municipal service buildings is particularly appropriate: it avoids reliance on external private actors and eliminates the risks associated with contractual interruptions or rental costs incompatible with educational budgets. Accordingly, this case study recommends deployment on service buildings, and more specifically on those owned by the municipality. Each region and each stakeholder has specific characteristics that the model cannot fully capture. The proposed model is an approximate representation built on a proximity logic; it does not claim to be a universal solution but a starting point for discussion

Urban Limits as Design Constraints: Identifying Suitable Locations for Distributed, Photovoltaic-Powered Servers

Workshop on Computing within Limits, June 23-25, 2026, ,

Table 3: Qualitative analysis of selected location typologies.

Service building Single-family housing Multi-family housing Parking area

Single-family housing Multi-family housing Service buildings Industrial buildings

Network

Energy

Structure

Number of owners

Nature of ownership

1 1 1 1

1 1 1 1

1 1 1 1

1 0 0 0.5

1 0 0.5 0.5

Sport buildings Parking Streetlight Polluted site

100 %

(11 : 00)

Numeric demand coverage

0% 100 %

(12 : 00)

0% 100 %

(13 : 00)

0% 100 %

(14 : 00)

0% 100 %

0%

(15 : 00) 1

2

3

4

Distance (km) Figure 5: Cumulative digital demand coverage as a function of the maximum distance to demand, shown for each energyfeasible location typology and for each peak solar hour (11:00–15:00 UTC+1). The filled circle on each curve marks the minimum distance at which that typology achieves full demand coverage.

with the potential stakeholders. They hold the detailed knowledge of needs and constraints that no modeled framework can generalize, and dialogue with them is needed to adjust the locations to be deployed.

5

Discussion on limitation of this work

Our methodology has a scope and a set of conditions that should be made explicit. It follows the spirit of LIMITS by treating digital infrastructure not as an abstract optimisation problem but as a socio-technical object embedded in finite urban, material, and ecological conditions. It works within urban limits by relying only on existing infrastructures and already anthropized surfaces, by integrating structural, social, and energy constraints, and by seeking proportionality between digital demand and the computing capacity actually deployed. A first limitation is the strong dependence on available data. In France, most relevant data come from government sources, which generally ensures accessibility and reliability. In other contexts, such data may be restricted or absent. International open data sources such as OpenStreetMap partly compensate, but their quality and completeness depend on user contributions and remain uneven. The methodology also depends on the territorial context in which it is applied: data availability, regulatory frameworks, urban morphology, ownership structures, and local energy resources all shape its transferability. It should therefore be read as a situated approach, to be locally adapted rather than directly replicated. A second limitation concerns the spatial assumptions of the model, which presupposes that rooftops and other anthropized surfaces can be mobilized. Such spaces may already host other technical devices – solar panels, ventilation systems, or equipment needed for the building’s normal operation – or support uses that contribute to urban sustainability, such as greening, urban agriculture, or shared gardens. As the paper notes, deploying computing capacity on existing infrastructures may conflict with the primary functions of places that already serve social, residential, or collective purposes. A third limitation is that our analysis considers the pedagogical benefits of LLM use without examining the full chain of involved actors. Teachers are a first essential group: integration requires pedagogical appropriation, adaptation of practices, and reflection on the practical conditions of classroom use. Citizens and residents are directly concerned as well, since the installation of computing units may create local nuisances such as noise, heat, or conflicts of use. Public authorities and urban stakeholders face economic constraints linked to the financing, deployment, maintenance, and renewal of these infrastructures, which is a major obstacle to implementation. As the paper also recalls, solar intermittency increases maintenance needs and limits continuous operation. Our study is limited to a single digital demand of a pedagogical nature, defined through a specific scenario. This framing makes the methodology testable but restricts the scope of the results. Extending the approach to other urban needs and other actors — housing providers, municipalities, citizens, public services, associations,

Workshop on Computing within Limits, June 23-25, 2026, ,

technical operators — would require treating these infrastructures as supporting a broader community of uses and energy flows [93]. Such an extension would change the management and governance questions: the challenge would no longer be only to identify a technically feasible infrastructure, but also to determine how access, contribution, maintenance, and decision-making are distributed among actors with potentially divergent interests. Territorial equity, socio-spatial justice, and collective governance would then become central to the long-term durability of this form of local digital sovereignty. The paper already integrates some dimensions of acceptability and equity but does not yet fully address these governance issues.

6

Conclusion and perspectives

Our work highlights a central insight regarding the deployment of digital technologies within the urban territory: a frugal digital service is not merely a matter of efficiency, but a matter of situated adequacy between needs, existing infrastructures, and local resources. By relying exclusively on already developed locations and on local photovoltaic production, we show that it is possible to support a digital service while respecting the environmental and social limits of the territory. The approach we adopted sheds light on the types of urban infrastructures that are most relevant for sustaining a distributed energy demand, both in terms of spatial availability, energy production capacity, and compliance with the social and structural constraints of the territory. Among the sixteen typologies initially identified, only a subset displayed sufficient capacity to contribute meaningfully to powering servers. Their use nevertheless depends on practical, institutional, and territorial considerations, especially when the objective is to maintain a local deployment strategy. The intrinsic robustness of the proposed framework stems from three complementary factors. First, the multiplicity of infrastructures—ranging from rooftops to streetlights—ensures wide spatial coverage and redundancy. Second, adherence to territorial constraints, including structural, social, and local digital demand considerations, anchors the deployment in realistic and acceptable urban contexts. Third, leveraging solar energy as a local resource reduces dependence on external energy supplies. Together, these elements prioritize robustness over peak performance, offering a sustainable digital service capable of withstanding a highly uncertain world marked by energy disruptions, conflicts, and other emerging risks. Beyond the experimental scope, this work highlights the importance of systematically integrating energy, spatial, and social criteria into the planning of emerging digital infrastructures. It also shows that distributed models supported by local photovoltaic resources and by the reuse of existing buildings constitute a promising avenue for reconciling digital usage with the reduction of associated environmental impacts. Such an approach opens the way to deployment strategies that are more robust, more sustainable, and more closely aligned with urban dynamics. In light of these results, several avenues for future research emerge. Applying this strategy to traditional studies on the optimization of computing-unit deployment (see Section 2) would ground the analysis in a realistic urban context while clarifying

Chikhaoui et al.

the types of digital services whose quality-of-service requirements could be met under such conditions. To avoid rebound effects, the evaluation metrics used in the literature also need to be reconsidered, so that they reward properties associated with sustainability rather than optimization alone. To avoid presenting any single solution as normative, results should be reported in a nuanced form – for example as a Pareto front [31] – that makes visible the trade-offs of prioritizing one sustainability metric over another. Another research avenue concerns the methodology’s ability to withstand urban change. Applying it to the same area at two different points in time would make it possible to evaluate the long-term relevance of an initial deployment. Finally, as mentioned earlier in the paper, future research directions should rely on active dialogue with urban stakeholders to identify desirable digital uses together with their needs and constraints, and to quantify them. Exploratory interviews [17] are one suitable instrument and allow different types of digital uses to be distinguished at an early stage. The relevance of the identified needs should then be assessed against a sustainability frame, drawing for instance on [76], which considers computing through planetary boundaries (natural resources, energy, climate), or on [49], which favors systems able to absorb fluctuations over those optimized for efficiency. Because each region and each stakeholder has specific characteristics, the proposed model – built on a proximity logic – is only an approximate representation and is not intended as a universal solution; it is a starting point for discussion with the actors who hold the detailed knowledge of needs and constraints that modeled frameworks can hardly capture.

Acknowledgments This work has been funded by the CNRS-MITI under the Explainer project grant. The authors used Generative AI tool for language editing and grammar checking. All technical content and results were produced and verified by the authors. The authors would like to thank all reviewers for their valuable comments and suggestions, which helped us to improve the quality of the paper.

References [1] Nasir Abbas, Yan Zhang, Amir Taherkordi, and Tor Skeie. 2017. Mobile edge computing: A survey. IEEE Internet of Things Journal 5, 1 (2017), 450–465. [2] M Jamil Ahmad and GN Tiwari. 2011. Solar radiation models—A review. International Journal of Energy Research 35, 4 (2011), 271–290. [3] Blake Alcott. 2005. Jevons’ paradox. Ecological economics 54, 1 (2005), 9–21. [4] Aleksandar Anastasovski, Maria Beatrice Andreucci, József Kádár, and Marco Delli Paoli. 2024. Energy storage in urban areas: the role of energy storage facilities, a review. Energies 17, 5 (2024), 1117. [5] Atakan Aral, Vincenzo De Maio, and Ivona Brandic. 2021. Ares: Reliable and sustainable edge provisioning for wireless sensor networks. IEEE Transactions on Sustainable Computing 7, 4 (2021), 761–773. [6] Atakan Aral, Vincenzo De Maio, and Ivona Brandic. 2021. ARES: Reliable and sustainable edge provisioning for wireless sensor networks. IEEE Transactions on Sustainable Computing 7, 4 (2021), 761–773. [7] Patricia Arroba, Rajkumar Buyya, Román Cárdenas, José L Risco-Martín, and José M Moya. 2024. Sustainable edge computing: Challenges and future directions. Software: Practice and Experience 54, 11 (2024), 2272–2296. [8] Ali Asghari and Mohammad Karim Sohrabi. 2024. Server placement in mobile cloud computing: A comprehensive survey for edge computing, fog computing and cloudlet. Computer Science Review 51 (2024), 100616. [9] Darren M Bagnall and Matt Boreland. 2008. Photovoltaic technologies. Energy Policy 36, 12 (2008), 4390–4396.

Urban Limits as Design Constraints: Identifying Suitable Locations for Distributed, Photovoltaic-Powered Servers

[10] Oliver Bates and Adrian Friday. 2018. Intangible commodities with free delivery: finding the limit in digitally mediated e-commerce and workforce injustice.. In Proceedings of the 2018 Workshop on Computing within Limits (Toronto, Ontario, Canada) (LIMITS ’18). Association for Computing Machinery, New York, NY, USA, Article 9, 8 pages. doi:10.1145/3232617.3232622 [11] Spie Batignolles. 2024. COASIS : solution conçue pour lutter contre les îlots de chaleur. Spie Batignolles. https://www.spiebatignolles.fr/coasissolution-concue-par-spie-batignolles-paysage-pour-lutter-contre-les-ilotsde-chaleur/ Consulté sur Spie Batignolles. [12] Emiliano Bellini. 2024. Un mur antibruit photovoltaïque avec un design en zigzag. PV Magazine France. https://www.pv-magazine.fr/2024/11/13/unmur-antibruit-photovoltaique-avec-un-design-en-zigzag/ Consulté sur PV Magazine. [13] Amira Rayane Benamer, Khaled Boussetta, and Nejib Ben Hadj-Alouane. 2021. A Genetic Algorithm for the Placement of Latency-Sensitive Multiplayer Game Servers in the Fog. In 2021 IEEE Global Communications Conference (GLOBECOM). 1–6. doi:10.1109/GLOBECOM46510.2021.9685952 [14] Luke Bennett. 2016. How does law make place? Localisation, translocalisation and thing-law at the world’s first factory. Geoforum 74 (2016), 182–191. [15] Salil Bharany, Sandeep Sharma, Osamah Ibrahim Khalaf, Ghaida Muttashar Abdulsahib, Abeer S Al Humaimeedy, Theyazn HH Aldhyani, Mashael Maashi, and Hasan Alkahtani. 2022. A systematic survey on energy-efficient techniques in sustainable cloud computing. Sustainability 14, 10 (2022), 6256. [16] Janusz Bialek. 1996. Tracing the flow of electricity. IEE Proceedings-Generation, Transmission and Distribution 143, 4 (1996), 313–320. [17] Alain Blanchet. 2007. L’enquête et ses méthodes: l’entretien. Armand Colin. [18] Geoff Boeing. 2018. Measuring the complexity of urban form and design. Urban Design International 23, 4 (2018), 281–292. [19] Gro Harlem Brundtland. 1987. Our common future world commission on environment and developement. (1987). [20] Bin Cao, Shanshan Fan, Jianwei Zhao, Shan Tian, Zihao Zheng, Yanlong Yan, and Peng Yang. 2021. Large-Scale Many-Objective Deployment Optimization of Edge Servers. IEEE Transactions on Intelligent Transportation Systems 22, 6 (June 2021), 3841 to 3849. doi:10.1109/tits.2021.3059455 [21] Walter Castelnovo, Gianluca Misuraca, and Alberto Savoldelli. 2016. Smart cities governance: The need for a holistic approach to assessing urban participatory policy making. Social Science Computer Review 34, 6 (2016), 724–739. [22] Le Chang, Xia Deng, Jianping Pan, and Yun Zhang. 2021. Edge server placement for vehicular ad hoc networks in metropolitans. IEEE Internet of Things Journal 9, 2 (2021), 1575–1590. [23] G Chaoui, R Yaagoubi, and M Mastere. 2025. Integrating geospatial technologies and multi-criteria decision analysis for sustainable and resilient urban planning. Chall. Sustain 13, 1 (2025), 122–134. [24] Yuanyi Chen, Yihao Lin, Zengwei Zheng, Peng Yu, Jiaxing Shen, and Minyi Guo. 2021. Preference-aware edge server placement in the internet of things. IEEE Internet of Things Journal 9, 2 (2021), 1289–1299. [25] Colas. [n. d.]. Mise en service de deux pistes cyclables photovoltaïques Wattway aux Pays-Bas. Colas — Communiqué de presse. https://www.colas.com/fr/communiques-de-presse/mise-en-service-de-deuxpistes-cyclables-photovoltaiques-wattway-aux-pays-bas Consulté sur le site officiel de Colas. [26] Colas Group. 2026. Wattway by Colas. https://www.wattwaybycolas.com/ Consulté le 27 février 2026. [27] Javier G. Corripio. 2023. insolation. https://pypi.org/project/insolation/ Python library for computing insolation over inclined surfaces or complex terrain. [28] Javier G Corripio. 2003. Vectorial algebra algorithms for calculating terrain parameters from DEMs and solar radiation modelling in mountainous terrain. International Journal of Geographical Information Science 17, 1 (2003), 1–23. [29] Marcos De Melo da Silva, Abdoulaye Gamatié, Gilles Sassatelli, Michael Poss, and Michel Robert. 2022. Optimization of data and energy migrations in mini data centers for carbon-neutral computing. IEEE Transactions on Sustainable Computing 8, 1 (2022), 68–81. [30] Marcos de Melo da Silva, Abdoulaye Gamatié, Gilles Sassatelli, Michael Poss, and Michel Robert. 2023. Optimization of Data and Energy Migrations in Mini Data Centers for Carbon-Neutral Computing. IEEE Trans. Sustain. Comput. 8, 1 (2023), 68–81. doi:10.1109/TSUSC.2022.3197090 [31] Kalyanmoy Deb. 2011. Multi-objective optimisation using evolutionary algorithms: an introduction. In Multi-objective evolutionary optimisation for product design and manufacturing. Springer, 3–34. [32] Gilles Desthieux, Claudio Carneiro, Reto Camponovo, Pierre Ineichen, Eugenio Morello, Anthony Boulmier, Nabil Abdennadher, Sébastien Dervey, and Christoph Ellert. 2018. Solar energy potential assessment on rooftops and facades in large built environments based on lidar data, image processing, and cloud computing. Methodological background, application, and validation in geneva (solar cadaster). Frontiers in Built Environment 4 (2018), 14. [33] Eleanor Jane Dowds and Fatme El-Saghir. 2021. Utilising waste heat from Edge-computing Micro Data Centres: Financial and Environmental synergies, Opportunities, and Business Models.

Workshop on Computing within Limits, June 23-25, 2026, ,

[34] Loïs Elziere. [n. d.]. Des panneaux solaires installés sur les rails du métro. Made in Marseille. https://madeinmarseille.net/actualite/173607-panneaux-solairesrails-metro-rose/ Consulté sur Made in Marseille. [35] Göran Finnveden, George Kamiya, Vlad C Coroamă, Pernilla Bergmark, and Reid Lifset. 2025. Assessing environmental impacts of digitalization: A special issue. 1042–1047 pages. [36] Manuel G Forero and HJ Herrera-Suárez. 2023. ChatGPT in the classroom: boon or bane for physics students’ academic performance? arXiv preprint arXiv:2312.02422 (2023). [37] Jay W Forrester. 1970. Urban dynamics. IMR; Industrial Management Review (pre-1986) 11, 3 (1970), 67. [38] French Republic. 2024. Article L111-19-1 du Code de l’urbanisme. République française. https://www.legifrance.gouv.fr Code de l’urbanisme, partie législative. [39] French Republic. 2024. Article R111-25-17 du Code de l’urbanisme. République française. https://www.legifrance.gouv.fr Code de l’urbanisme, partie réglementaire. [40] Abdoulaye Gamatié, Thomas Leduc, Daniel Siret, Gilles Sassatelli, and Michel Robert. 2023. A model-based approach to addressing energy demand in sustainable urban systems. Sustainable Computing: Informatics and Systems 37 (2023), 100844. [41] Apurva Goel, Snehal Masurkar, and Girish R Pathade. 2024. An overview of digital transformation and environmental sustainability: Threats, opportunities, and solutions. Sustainability 16, 24 (2024), 11079. [42] Government of France. 2023. Article L.632-2 du Code du patrimoine. République française. https://questions.assemblee-nationale.fr/q16/16-7059QE.htm Texte cité dans la réponse à la question écrite n°7059, Assemblée nationale. [43] Government of France. 2026. Code de la construction et de l’habitation. Légifrance. https://www.legifrance.gouv.fr/loda/id/LEGISCTA000006189287 Consulté sur Légifrance, version en vigueur. [44] Government of France. 2026. Code de l’environnement. Légifrance. https://www. legifrance.gouv.fr/loda/id/LEGISCTA000032445080 Consulté sur Légifrance, version en vigueur. [45] Government of France. 2026. Code de l’urbanisme. Légifrance. https://www. legifrance.gouv.fr/codes/texte_lc/LEGITEXT000006074075/2023-09-30 Consulté sur Légifrance, version en vigueur. [46] Government of France. 2026. Code de l’énergie. Légifrance. https://www. legifrance.gouv.fr/codes/texte_lc/LEGITEXT000023983208/2023-11-10 Consulté sur Légifrance, version en vigueur. [47] Christian Gueymard. 1989. A two-band model for the calculation of clear sky solar irradiance, illuminance, and photosynthetically active radiation at the earth’s surface. Solar Energy 43, 5 (1989), 253–265. [48] Liwen Guo, Jiahong Wu, I-Min Hsieh, Cheng-Hua Huang, Bo Li, Chiming Jao, Vinson Lin, Robert Yuan, Xianguang Tan, Xiaojin Fan, et al. 2021. Advanced Single-phase Passive Immersion Cooling Solution with Natural Convection for Outdoor Edge Servers. In 2021 20th IEEE Intersociety Conference on Thermal and Thermomechanical Phenomena in Electronic Systems (iTherm). IEEE, 417–422. [49] Olivier Hamant. 2023. Antidote au culte de la performance: la robustesse du vivant. Gallimard. [50] Kate Hosford, Meghan Winters, Kadia Saint-Onge, Nazeem Muhajarine, and Lise Gauvin. 2024. Acceptability of built environment interventions to support active travel in 17 Canadian metropolitan areas: findings from the THEPA study. Sustainable Transport and Livability 1, 1 (2024), 2314024. [51] Lara Houston, Ann Light, and Cassie Thornton. 2022. The richness of designing for eco-social change. In LIMITS’22: Workshop on Computing within Limits. 21–22. [52] Pengfei Hu, Sahraoui Dhelim, Huansheng Ning, and Tie Qiu. 2017. Survey on fog computing: architecture, key technologies, applications and open issues. Journal of network and computer applications 98 (2017), 27–42. [53] Pierre Ineichen and Richard Perez. 2002. A new airmass independent formulation for the Linke turbidity coefficient. Solar Energy 73, 3 (2002), 151–157. [54] Adam R Jensen, Kevin S Anderson, William F Holmgren, Mark A Mikofski, Clifford W Hansen, Leland J Boeman, and Roel Loonen. 2023. pvlib iotools—Opensource Python functions for seamless access to solar irradiance data. Solar Energy 266 (2023), 112092. [55] Kelsey Jordahl, Joris Van den Bossche, Martin Fleischmann, Jacob Wasserman, James McBride, Jeffrey Gerard, Jeff Tratner, Matthew Perry, Adrian Garcia Badaracco, Carson Farmer, Geir Arne Hjelle, Alan D. Snow, Micah Cochran, Sean Gillies, Lucas Culbertson, Matt Bartos, Nick Eubank, maxalbert, Aleksey Bilogur, Sergio Rey, Christopher Ren, Dani Arribas-Bel, Leah Wasser, Levi John Wolf, Martin Journois, Joshua Wilson, Adam Greenhall, Chris Holdgraf, Filipe, and François Leblanc. 2020. geopandas/geopandas: v0.8.1. doi:10.5281/zenodo.3946761 [56] JPEE. [n. d.]. Inaugurations — Projets d’énergie renouvelable. JPEE. https: //www.jpee.fr/inauguration/ Consulté sur JPEE. [57] Jessica Jungell-Michelsson and Pasi Heikkurinen. 2022. Sufficiency: A systematic literature review. Ecological Economics 195 (2022), 107380. [58] Sandeep Kautish and Dipesh Gurung. 2025. Advancing Sustainable Computing: A Systematic Literature Review of Software, Hardware, and Algorithmic

Workshop on Computing within Limits, June 23-25, 2026, ,

Innovations. ICCK Transactions on Sustainable Computing 1, 1 (2025), 1–19. [59] Gregory Kestin, Kelly Miller, Anna Klales, Timothy Milbourne, and Gregorio Ponti. 2024. AI tutoring outperforms active learning. (2024). [60] Lars Krupp, Steffen Steinert, Maximilian Kiefer-Emmanouilidis, Karina E Avila, Paul Lukowicz, Jochen Kuhn, Stefan Küchemann, and Jakob Karolus. 2023. Unreflected acceptance–investigating the negative consequences of ChatGPTassisted problem solving in physics education. arXiv preprint arXiv:2309.03087 (2023). [61] Moyukh Laha, Suraj Kamble, and Raja Datta. 2020. Edge nodes placement in 5G enabled urban vehicular networks: A centrality-based approach. In 2020 National Conference on Communications (NCC). IEEE, 1–6. [62] Tero Lähderanta, Teemu Leppänen, Leena Ruha, Lauri Lovén, Erkki Harjula, Mika Ylianttila, Jukka Riekki, and Mikko J Sillanpää. 2021. Edge computing server placement with capacitated location allocation. J. Parallel and Distrib. Comput. 153 (2021), 130–149. [63] Mireille Lefèvre, Armel Oumbe, Philippe Blanc, Bella Espinar, Benoît Gschwind, Zhipeng Qu, Lucien Wald, Marion Schroedter-Homscheidt, Carsten Hoyer-Klick, Antti Arola, et al. 2013. McClear: a new model estimating downwelling solar radiation at ground level in clear-sky conditions. Atmospheric Measurement Techniques 6, 9 (2013), 2403–2418. [64] Fonroche Lighting. [n. d.]. 50 000 lampadaires solaires pour éclairer le Sénégal. Fonroche Lighting. https://www.fonroche-lighting.com/fr/fra/projets/projetsgouvernementaux/50-000-lampadaires-solaires-pour-eclairer-le-senegal Consulté sur Fonroche Lighting. [65] Haojiang Liu, Yuanzhe Li, and Shangguang Wang. 2022. Request scheduling combined with load balancing in mobile-edge computing. IEEE Internet of Things Journal 9, 21 (2022), 20841–20852. [66] Lauri Lovén, Tero Lähderanta, Leena Ruha, Teemu Leppänen, Ella Peltonen, Jukka Riekki, and Mikko J Sillanpää. 2020. Scaling up an edge server deployment. In 2020 IEEE International Conference on Pervasive Computing and Communications Workshops (PerCom Workshops). IEEE, 1–7. [67] Weisheng Lu, Jinfeng Lou, Benjamin Kwaku Ababio, Ray Y Zhong, Zhikang Bao, Xiao Li, and Fan Xue. 2024. Digital technologies for construction sustainability: Status quo, challenges, and future prospects. Npj Materials Sustainability 2, 1 (2024), 10. [68] Alexandra Sasha Luccioni, Emma Strubell, and Kate Crawford. 2025. From efficiency gains to rebound effects: The problem of jevons’ paradox in AI’s polarized environmental debate. In Proceedings of the 2025 ACM conference on fairness, accountability, and transparency. 76–88. [69] Oluwashina Madamori, Esther Max-Onakpoya, Gregory D Erhardt, and Corey E Baker. 2021. Enabling opportunistic low-cost smart cities by using tactical edge node placement. In 2021 16th Annual Conference on Wireless On-demand Network Systems and Services Conference (WONS). IEEE, 1–8. [70] Oluwashina Madamori, Esther Max-Onakpoya, Gregory D. Erhardt, and Corey E. Baker. 2021. A Latency-Defined Edge Node Placement Scheme for Opportunistic Smart Cities. In 2021 IEEE International Conference on Pervasive Computing and Communications Workshops and other Affiliated Events (PerCom Workshops). 142–147. doi:10.1109/PerComWorkshops51409.2021.9430977 [71] Made in Marseille. 2026. La RTM teste des rails photovoltaïques sur les tronçons aériens du métro. https://madeinmarseille.net/environnement/136761-la-rtmteste-des-rails-photovoltaiques-sur-les-troncons-aeriens-du-metro/ Consulté le 27 février 2026. [72] Robson Mandishekwa and Enard Mutenheri. 2020. Mining-induced displacement and resettlement: An analytical review. Ghana Journal of Development Studies 17, 1 (2020), 114–140. [73] Ani Matei and Mădălina Cocos, atu. 2024. Artificial internet of things, sensorbased digital twin urban computing vision algorithms, and blockchain cloud networks in sustainable smart city administration. Sustainability 16, 16 (2024), 6749. [74] Marcelle C McManus. 2012. Environmental consequences of the use of batteries in low carbon systems: The impact of battery production. Applied Energy 93 (2012), 288–295. [75] Carla Mouradian, Diala Naboulsi, Sami Yangui, Roch H Glitho, Monique J Morrow, and Paul A Polakos. 2017. A comprehensive survey on fog computing: State-of-the-art and research challenges. IEEE communications surveys & tutorials 20, 1 (2017), 416–464. [76] Bonnie Nardi, Bill Tomlinson, Donald J Patterson, Jay Chen, Daniel Pargman, Barath Raghavan, and Birgit Penzenstadler. 2018. Computing within limits. Commun. ACM 61, 10 (2018), 86–93. [77] Wacuka M Ngata, Noman Bashir, Michelle Westerlaken, Laurent Liote, Yasra Chandio, and Elsa Olivetti. 2025. The Cloud next door: investigating the environmental and socioeconomic strain of datacenters on local communities. In Proceedings of the ACM SIGCAS/SIGCHI Conference on Computing and Sustainable Societies. 769–774. [78] Rémi Paccou and Pr Fons Wijnhoven. 2024. Artificial Intelligence and Electricity: A System Dynamics Approach. Technical Report. Schneider Electric. https: //hal.science/hal-05015754

Chikhaoui et al.

[79] Eleni Papadonikolaki, Ilias Krystallis, and Bethan Morgan. 2022. Digital technologies in built environment projects: review and future directions. Project Management Journal 53, 5 (2022), 501–519. [80] Ravi Parthiban and P Ponnambalam. 2022. An enhancement of the solar panel efficiency: a comprehensive review. Frontiers in Energy Research 10 (2022), 937155. [81] Céline Péréa, Jessica Gérard, and Julien de Benedittis. 2023. Digital sobriety: From awareness of the negative impacts of IT usages to degrowth technology at work. Technological Forecasting and Social Change 194 (2023), 122670. [82] Leonardo MO Queiroz, Marcio A Roselli, Celso Cavellucci, and Christiano Lyra. 2012. Energy losses estimation in power distribution systems. IEEE Transactions on Power Systems 27, 4 (2012), 1879–1887. [83] Mohammad MH Raihan, Sujoy Subroto, Nashit Chowdhury, Katharina Koch, Erin Ruttan, and Tanvir C Turin. 2025. Dimensions and barriers for digital (in) equity and digital divide: a systematic integrative review. Digital Transformation and Society 4, 2 (2025), 111–127. [84] Arun Pratap Singh Rathod, Poornima Mittal, and Brijesh Kumar. 2016. Analysis of factors affecting the solar radiation received by any region. In 2016 International Conference on Emerging Trends in Communication Technologies (ETCT). IEEE, 1–4. [85] Tim Reddel and Geoff Woolcock. 2004. From consultation to participatory governance? A critical review of citizen engagement strategies in Queensland. Australian Journal of Public Administration 63, 3 (2004), 75–87. [86] Gilles Sassatelli, Abdoulaye Gamatié, and Michel Robert. 2016. Data Processing System with Energy Transfer. https://hal-lirmm.ccsd.cnrs.fr/lirmm02145436 Related patent documents:FR3050295 (A1) FR3050295 (B1) CA3020280 (A1) CN109074285 (A) EP3443457 (A1) JP2019514329 (A) KR20180134355 (A) RU2018139497 (A) RU2018139497 (A3) US2019163541 (A1) WO2017178571 (A1). [87] Neil Selwyn. 2024. Digital degrowth: Toward radically sustainable education technology. Learning, Media and Technology 49, 2 (2024), 186–199. [88] Rehab Shahin, Ali A El-Moursy, Sherif M Saif, Hazem M Abbas, and Salwa M Nassar. 2021. Fog node optimum placement and configuration technique for vanets. In 2020 International Conference on Communications, Signal Processing, and Their Applications (ICCSPA). IEEE, 1–6. [89] Rehab Shahin, Sherif M Saif, Ali A El-Moursy, Hazem M Abbas, and Salwa M Nassar. 2023. Fog-ROCL: A Fog based RSU Optimum Configuration and Localization in VANETs. Pervasive and Mobile Computing 94 (2023), 101807. [90] Yanling Shao, Zhen Shen, Siliang Gong, and Hanyao Huang. 2022. Cost-Aware Placement Optimization of Edge Servers for IoT Services in Wireless Metropolitan Area Networks. Wireless Communications and Mobile Computing 2022, 1 (2022), 8936576. [91] Daniel A Shtob, Jordan Fox, and Patrick Trent Greiner. 2025. Planning for the complexity and uncertainty of urban socio-environmental futures. Nature Cities 2, 3 (2025), 187–197. [92] M Six Silberman. 2015. Information systems for the age of consequences. First Monday (2015). [93] Robert Soden, Matt Ratto, G Arno Verhoeven, and Bart Simon. 2022. Photovoltaic Imagination: Solar Strategies for Community Integrated Research and Graduate Training. In Eighth Workshop on Computing within Limits. [94] Jacob T Stid, Sam Zipper, Hanna V Szydlowski, James Bingaman, Annick Anctil, and Anthony D Kendall. [n. d.]. Current Knowledge on Low-Impact GroundMounted Solar Siting, Construction, and Installation Practices. ([n. d.]). [95] Christian Stöhr, Amy Wanyu Ou, and Hans Malmström. 2024. Perceptions and usage of AI chatbots among students in higher education across genders, academic levels and fields of study. Computers and Education: Artificial Intelligence 7 (2024), 100259. [96] Emily Talen. 2012. City rules: How regulations affect urban form. Island Press. [97] Puay Yok Tan and Rosita Samsudin. 2017. Effects of spatial scale on assessment of spatial equity of urban park provision. Landscape and Urban Planning 158 (2017), 139–154. [98] Ciel & Terre. [n. d.]. Premier projet photovoltaïque flottant d’ampleur en Alsace. Ciel & Terre. https://ciel-et-terre.net/fr/la-commune-de-leutenheim-accueillele-premier-projet-photovoltaique-flottant-dampleur-en-alsace-fruit-deletroite-collaboration-entre-generale-du-solaire-et-ciel-terre/ Consulté sur Ciel & Terre. [99] Sebastian Theis. 2025. From local to global: leveraging localized environmental actions for scalable sustainability effects. PLOS Sustainability and Transformation 4, 7 (2025), e0000185. [100] Klervie Toczé, Maël Madon, Muriel Garcia, and Patricia Lago. 2022. The dark side of cloud and edge computing: an exploratory study. In LIMITS’22: Workshop on Computing within Limits. [101] Bill Tomlinson, Eli Blevis, Bonnie Nardi, Donald J Patterson, M SIX Silberman, and Yue Pan. 2013. Collapse informatics and practice: Theory, method, and design. ACM Transactions on Computer-Human Interaction (TOCHI) 20, 4 (2013), 1–26. [102] Tanya Tsui, David Peck, Bob Geldermans, and Arjan Van Timmeren. 2020. The role of urban manufacturing for a circular economy in cities. Sustainability 13, 1 (2020), 23.

Urban Limits as Design Constraints: Identifying Suitable Locations for Distributed, Photovoltaic-Powered Servers

[103] Urbaneo. [n. d.]. Abribus Hélios solaire. Urbaneo. https://www.urbaneo.eu/ mobiliers-urbains/abribus-helios-solaire/ Consulté sur Urbaneo. [104] Shangguang Wang, Yan Guo, Ning Zhang, Peng Yang, Ao Zhou, and Xuemin Shen. 2019. Delay-aware microservice coordination in mobile edge computing: A reinforcement learning approach. IEEE Transactions on Mobile Computing 20, 3 (2019), 939–951. [105] Alan Wiig. 2015. IBM’s smart city as techno-utopian policy mobility. City 19, 2-3 (2015), 258–273. [106] Wikipedia. [n. d.]. Route solaire. Wikipédia. https://fr.wikipedia.org/wiki/ Route_solaire Consulté sur Wikipédia. [107] Eric Williams. 2011. Environmental effects of information and communications technologies. nature 479, 7373 (2011), 354–358. [108] Mohammad Yamin. 2019. Information technologies of 21st century and their impact on the society. International Journal of Information Technology 11, 4 (2019), 759–766. [109] Sing Chen Yeo, Jacinda Tan, June C Lo, Michael WL Chee, and Joshua J Gooley. 2020. Associations of time spent on homework or studying with nocturnal sleep

Workshop on Computing within Limits, June 23-25, 2026, ,

behavior and depression symptoms in adolescents from Singapore. Sleep Health 6, 6 (2020), 758–766. [110] Ashkan Yousefpour, Caleb Fung, Tam Nguyen, Krishna Kadiyala, Fatemeh Jalali, Amirreza Niakanlahiji, Jian Kong, and Jason P Jue. 2019. All one needs to know about fog computing and related edge computing paradigms: A complete survey. Journal of Systems Architecture 98 (2019), 289–330. [111] ZGSM. [n. d.]. Solar-powered traffic light systems in Portugal. ZGSM — Case Study. https://fr.zgsm-china.com/case-study/solar-powered-traffic-lightsystems-in-portugal.html Consulté sur ZGSM. [112] Dongyu Zhu, Tao Chen, Na Zhen, and Ruiqing Niu. 2020. Monitoring the effects of open-pit mining on the eco-environment using a moving window-based remote sensing ecological index. Environmental Science and Pollution Research 27, 13 (2020), 15716–15728.

Received ; revised ; revised