A Periodic Space of Distributed Computing: Vision & Framework

arXiv:2604.12259v1 [cs.DC] 14 Apr 2026

A Periodic Space of Distributed Computing: Vision & Framework Mohsen Amini Salehi∗

Adel N. Toosi∗

Hai Duc Nguyen∗

University of North Texas Denton, TX, USA [email protected]

The University of Melbourne Melbourne, Australia [email protected]

Argonne National Laboratory Lemont, IL, USA [email protected]

Murtaza Rangwala∗

Omer Rana

Tevfik Kosar

The University of Melbourne Melbourne, Australia [email protected]

Cardiff University Cardiff, UK [email protected]

University at Buffalo Buffalo, NY, USA [email protected]

Valeria Cardellini

Rajkumar Buyya

Tor Vergata University of Rome Rome, Italy [email protected]

The University of Melbourne Melbourne, Australia [email protected]

Abstract

tiers spanning devices, edge and fog layers, and large-scale cloud infrastructures, each characterized by distinct constraints and service abstractions [5]. This evolution has enabled computation to migrate fluidly across the continuum to satisfy application requirements for latency, privacy, and quality of service [6–8]. As a result, distributed computing systems are transitioning from an active research domain into critical global infrastructure [9, 10], forming the foundation for distributed intelligence [11], where AI systems operate collectively across large-scale environments. Understanding the dynamics of this emerging landscape is therefore essential, as it will underpin future intelligent systems and may ultimately shape the technological trajectory of the 21st century. As computing environments become increasingly pervasive and interconnected, distributed systems are approaching a fundamental paradigm shift in how they are designed, governed, and consumed. In this emerging landscape, computing resources are becoming a strategic asset, often compared to the “oil” of the 21st century. The rapid rise of infrastructure-intensive technologies such as generative AI and large language models (LLMs) [12], combined with the slowing of Moore’s Law as semiconductor scaling approaches physical limits [13], suggests that the coming decades will be defined by a profound bottleneck in computational capacity [14]. This shift introduces both opportunities and challenges. Limited access to advanced computing infrastructure, particularly in developing regions [15], risks widening the digital divide and constraining productivity, innovation, and global competitiveness [16]. At the same time, the rapid expansion of distributed systems and hyperscale data centers brings significant societal and environmental costs, including rising energy and water consumption, noise pollution, and an enlarged attack surface for cyber threats [17]. Together, these trends highlight the growing importance of rethinking how future distributed computing infrastructures are structured and managed. Understanding how distributed computing infrastructures should evolve in response to these pressures requires a careful examination of its characteristics, evolution, and ramifications. Yet this task is increasingly difficult as the distributed computing continuum continues to expand in both scale and complexity, with new frameworks, architectures, and abstractions emerging at a rapid pace,

Advances in networking and computing technologies throughout the early decades of the 21st century have transformed longstanding dreams of pervasive communication and computation into reality. These technologies now form a rapidly evolving and increasingly complex global infrastructure that will underpin the next aspiration of computing: supporting intelligent systems with human-level or even superhuman capabilities. We examine how today’s distributed computing landscape can evolve to meet the demands of future users, intelligent systems, and emerging application domains. We propose a “periodic framework” for characterizing the distributed computing landscape, inspired by the systematic structure and explanatory power of the “periodic table” in chemistry. This framework provides a structured way to describe, compare, and reason about the behaviors and design choices of different distributed computing solutions. Using this framework, we can identify patterns in key system properties, such as responsiveness and availability, across the distributed computing landscape. We also explain how the framework can help in predicting future trajectories in the field. Lastly, we synthesize insights from leading researchers worldwide regarding the desired properties, design principles, and implications of the emerging areas in the forthcoming distributed computing landscape and in relation to the periodic framework. Together, these perspectives shed light on the considerations that will shape the distributed computing landscape underpinning future intelligent systems.

1

Introduction

The first decades of the 21st century witnessed rapid progress toward realizing two long-standing aspirations of computing: pervasive communication and computation. Successive generations of networking technologies, the widespread adoption of Internetconnected devices, and more recently low-Earth-orbit satellite constellations [1, 2] have transformed connectivity from a regional capability into a truly global one. In parallel, hyperscale cloud platforms [3] and the proliferation of IoT systems [4] have reshaped computing into a continuum of interconnected and heterogeneous ∗ These authors contributed equally to this work.

1

L7: Agents

Microsoft 365

Smart Surveillance

L6: Application

TensorFlow

L5: Programming Models

Abstraction Levels

remaining flexible enough to accommodate emerging solutions as the distributed computing ecosystem continues to evolve. This framework arranges computing solutions along two orthogonal dimensions: tiers continuum on the horizontal axis, spanning computing infrastructure from the extreme local devices and edge, to centralized cloud and beyond; and the abstraction level on the vertical axis, capturing the progression from low-level hardware to high-level Agents AI services. By positioning existing solutions within this space, the framework provides a unified lens for explaining relationships among solutions, highlighting design trade-offs, and identifying unexplored or emerging regions that point toward future computing models. We note that the chemical periodic table is discrete, i.e., each element has a single position in the table that reflects its fundamental properties and relationships (e.g., atomic structure and chemical behavior). However, our Periodic Space represents a continuous space where a given solution may span over more than one tier and abstraction level.

Agentic AI Services

FaaS (e.g., ?)

L4 : Execution (Runtime) CaaS (e.g. K3s)

L3: Platform

CaaS (e.g. K8s) Quantum Platform

Containers (e.g. Docker) L2:Inf rastructure IaaS (e.g, VMs) L1: Hardw are (No Abstraction)

Arduino T1: Local (Device)

Jestson Nano T2: Edge

Bare Metal Servers T3: Fog

T4 : Cloud

T5: Sky

Tiers Continuum

Figure 1: A “Periodic Space” of distributed computing landscape, organized by Tiers (horizontal) and Abstractions (vertical). Representative solutions are positioned within the space. Gray boxes indicate futuristic or emerging solutions.

2.1.1 Tiers Continuum. To support applications with diverse requirements in performance, reliability, and data locality, computing infrastructure has naturally evolved across multiple physical layers. At one end of the spectrum are resource-constrained devices (e.g., IoT sensors and cameras) where computation occurs close to data sources. At the other end are hyperscale cloud data centers designed for large-scale analytics and elastic resource provisioning. These layers reflect fundamental trade-offs among proximity to data, resource capacity, and operational scope. Positioning distributed computing solutions along this dimension, based on where their underlying resources reside and the scale at which they operate, provides a natural way to reason about, compare, and understand their relationship with the underlying infrastructure. In practice, there are no strict boundaries between computation tiers, so we make the spectrum infinite and continuous to allow future expansion. However, for clearer presentation and understanding, we use commonly recognized tiers as reference anchors across this continuum. Local Device—Computation occurs at the same location where data is generated or user requests originate (e.g., on-device video analytics [19] and sensor data processing on wearables [20]). Edge—Computation is placed within the same local network or physical vicinity as the data source, such as servers deployed in factories [21] or mobile and energy-constrained systems [22, 23]. Fog (aka Cloudlet)—resources are provided by regional data centers with no energy constraint, some level of elasticity, and slightly more latency in compare to the edge tier [24]. Examples are campus clusters and ISP-operated mini–data centers that offer domainspecific services [25] . Cloud—resources come from large, widely distributed (public or private) datacenters and can be provisioned globally. Clouds are characterized by their elasticity, and large-scale computation for use cases such as big data analytics, and machine learning training. Major global public cloud providers are AWS, Azure, and GCP. Sky—resources extend beyond a single cloud to encompass crosscloud deployments [26]. More recently, this concept has been extended to computing infrastructures spanning terrestrial, aerial, and space-based platforms [27], such as LEO satellites and high-altitude platforms, and even extra-planetary computing infrastructures, such as planetary-scale and off-world data centers [28].

each introducing distinct assumptions and design trade-offs. This rapid evolution makes it challenging to form a unified view of the ecosystem or to reason systematically about its long-term trajectory. To address this challenge, we argue for a principled approach to organizing the computing landscape into a self-explanatory and predictive structure—one capable of revealing relationships, constraints, and system behaviors without requiring detailed knowledge of individual solutions. A similar intellectual breakthrough occurred with the development of the Chemical Periodic Table [18], which organized elements in a way that made their properties and relationships interpretable and even enabled predictions of undiscovered elements. Inspired by this idea, we seek an analogous organizing principle: a conceptual periodic space that provides an intuitive and systematic framework for characterizing existing and future solutions across the distributed computing landscape. Building on this principle, we propose a framework that captures the fundamental design constraints and system properties of distributed computing across the tiers continuum and levels of abstraction. We examine how this framework can describe recurring patterns and help reason about the behavior of existing and emerging solutions across the distributed computing landscape. To ground this perspective, we also incorporate insights from leading researchers working at the boundaries of the proposed periodic space. By providing a structured view of the evolving ecosystem, this work aims to inform researchers, technology leaders, and policymakers in shaping the next generation of distributed systems.

2 Periodic Framework of Distributed Systems 2.1 Periodic Space We propose the concept of a Periodic Space for distributed computing, illustrated in Figure 1. The central idea is to provide a conceptual framework that organizes the distributed computing landscape into a continuous and extensible space, enabling solutions to be characterized by their underlying constraints, capabilities, and levels of abstraction rather than by isolated taxonomies or niche properties. In doing so, the periodic space offers a systematic way to describe relationships among existing solutions while 2

2.1.2 Abstraction Levels. As computing resources expand across the tiers continuum, developing, deploying, and maintaining distributed applications has become increasingly complex. To manage this complexity, software layers introduce abstractions that hide infrastructure details such as hardware heterogeneity, resource placement, communication, and failure handling, allowing developers to focus on application logic rather than the mechanics of distribution. Higher levels of abstraction simplify development and deployment but reduce visibility into, and control over, the underlying resources. Similar to the tiers dimension, we model abstraction as an infinite and continuous spectrum, since infrastructure can be abstracted in many ways across computation, communication, and data management without clear boundaries between levels. Because abstractions emerge from developer needs, distributed solutions can be positioned along this spectrum according to which parts of their software stack are abstracted away: Bare Hardware (No Abstraction)—solutions expose raw hardware with no abstraction, such as bare metal servers in data centers, or naked single board computers at the edge (e.g., Raspberry Pi [29] or an Nvidia Jetson Nano [30]). Infrastructure—hardware complexity is abstracted through virtualization technologies such as virtual machines, containers, or emerging WebAssembly-based runtimes [31, 32]. Platform—environment dependencies are abstracted away, including operating systems, libraries, and runtime configuration. By shielding developers from low-level infrastructure management, the platform layer enables consistent deployment and execution across heterogeneous resources. Cloud middleware platforms (e.g., Kubernetes [33]) or open-source Infrastructure as Code (e.g., Terraform [34]) are categorized in this level. Execution (Runtime)—provides a managed execution environment for running applications via abstracting operational tasks such as resource management, elasticity, and fault-tolerance. Examples include serverless execution environments (e.g., KNative [35]) that execute code in response to events. Programming Models — abstract away complexities at application scale (i.e., cross-solution integration), including workflow orchestration (AWS Step Functions [36]), Quality of Service (QoS) guarantees (RBAM [37, 38]), end-to-end development and deployment (e.g., OaaS [39–41]). Application aka software as a service (SaaS)—in this level, a service provides the entire ready-to-use software application with minimal configuration (e.g., Microsoft 365 [42]). Agents—abstract away human effort on developing, deploying, and managing solutions, currently through autonomous or semiautonomous agentic services [43] that encapsulate application logic, decision-making, and action execution. Examples include autonomous data analysis agents [44], task-oriented copilots [45], and multi-agent systems [46].

2.2

the periodic-space framework, we enable structured comparison and obtain deeper insight into how placement across tiers and abstraction levels influences overall system behavior. In particular, this structure allows us to reason how a given solution is likely to behave as it moves along the periodic space.

Securit y & Trust w orthiness

Sustainabilit y AI- Support

Capacit y

AI- Native

Responsiveness

Controllabilit y

M ajor System Proper ties of the D istr ibuted Com puting L andscape

Availabilit y

Democratization

Inf rastructure Cost

Reliabilit y

Operational Cost

Interoperabilit y

Distributedness

Elastisit y Mobilit y

Simple

Sy stem

Abstractions Propert y

Low

Intensit y

High

Compound

Tiers

Figure 2: System properties across computing tiers (left to right: local → sky) and abstraction levels (bottom to top: hardware → agent), categorized as simple and compound; color indicates intensity relative to the periodic space. We identify a set of key system properties, summarized in Figure 2, with their definitions provided in Table 1. These properties often exhibit distinct, either uniform or non-uniform trends across the periodic space. We categorize the properties into two groups: Simple and Compound. Simple properties, e.g., responsiveness and capacity, typically exhibit more straightforward trends across the periodic space and are easier to intuitively represent and reason about as the space evolves. In contrast, Compound properties emerge from the interaction of multiple simple properties, hence, require multi-metric evaluation. For example, reliability depends on latency, failure rates, and the degree of distribution, and governance depends on access control and observability. As illustrated in Figure 2, we characterize each system property by highlighting its trends and variations across the periodic space. These trends and variations are visually encoded using intensitybased background coloring for each property. Through several examples below, we demonstrate how the periodic space can be used to characterize and reason about system properties across the distributed computing landscape. Example 1: Responsiveness. As a solution moves from lower, locality-oriented tiers (e.g., a container at the edge) to higher tiers (the same container deployed in the cloud), it typically experiences increased latency and reduced responsiveness due to greater physical and network distance from end users and data sources [47]. Example 2: Democratization and Ease-of-Use. When a solution moves up across abstraction levels, it becomes easier to develop,

Periodic Space for System Properties

Just as the periodic table of chemical elements conveys fundamental properties based on the position of each element, our proposed periodic space can serve as a lens for understanding and characterizing solutions and systems within the space with respect to a range of secondary dimensions, which we collectively refer to as “system properties”. By tracing trends of these properties across 3

Table 1: Key system properties of solutions in distributed systems; along with their definition and measurement (evaluation) metric. The last two columns show how each property is impacted by changes across the periodic space: ↑/↓ = increase/decrease as abstraction/tier increase, − = no impact, and ? = unknown, unclear or non-linear pattern★. Property

Definition

Common Measure- Abstraction ment Capacity Computational power a solution provides in terms of processing, Throughput (re− memory, storage, and network. quests/sec) Speed with which a solution processes and returns responses Latency (sec) Responsiveness ↓ to user/client requests. Capital Cost∗ The total software & hardware expenses to develop and establish Money ($) ↑ a solution. Operational Cost∗ The total economic expense to operate and run a solution. Money ($) ↓ The ability of a solution to adapt its resource allocation to load Load-Metric Curve Elasticity ↑ changes in order to maintain performance objectives. or Elasticity ratio Reliability The ability of a solution to provide correct and dependable Availability (%) ? services over time. Mobility The ability of a solution to work on mobile compute resources N/A − (e.g., smartphones or vehicular nodes) while maintaining uninterrupted services. The extent to which computation, data, and control are spread N/A Distributedness ↑ across multiple geographically and logically distinct nodes. The ability of a solution to operate across platforms, technolo- N/A ↑ Interoperability gies, and administrative domains. The degree to which a solution lowers barriers to access, devel- Learning Hours or ↑ Democratization opment, and deployment through ease of use, programmability, Time to Delivery and accessibility. Controllability ↓ The degree to which developers or users can directly configure, N/A manage, and influence infrastructure behavior and execution decisions. AI-Native AI-optimized solutions where AI is fundamentally integrated N/A ? into the system’s design and operation. AI-Support Solution that are built to efficiently support AI workloads. N/A ? The ability of a solution to minimize carbon emission while Carbon emission Sustainability‡ ↓ remaining effective over time. (𝐶𝑂 2𝑒) Security & Trust- The ability of a solution to protect data, operations, and users N/A ? worthiness against threats while ensuring integrity, confidentiality, and trustworthy behavior.

Tiers ↑ ↓ ↓ ↑ ↑ ↑ ↓

↓† ? ?

↓

↑ ↑ ↑ ?

★ For more insights, please refer to the website we developed for this purpose: https://hpcclab.github.io/periodic-table. ∗ These metrics are to be interpreted

from the service provider’s point of view. † This trend will show a slight increase as it moves from the Cloud tier to the Sky Tier. ‡ This metric is strictly considered from the lens of energy usage.

2.3

deploy, and maintain–reflecting increased democratization and easeof-use. However, this often comes at the cost of lower governance and trustworthiness, as control over underlying system behavior is increasingly delegated to the platform [48]. Example 3: Elasticity. This property exhibits clear trends across both tier and abstraction dimensions. Higher-level tiers, such as the cloud, naturally provide more elastic behavior than local or edge tiers due to their access to larger resource pools. In addition, introducing higher levels of software abstraction in the stack enables greater flexibility in resource management–allowing more elastic solutions in compare to deployments tightly coupled to nonabstracted hardware.

Periodic Space to Anticipate Future Trends

Beyond characterization, the periodic space also has an anticipative role: by observing how system properties evolve along its dimensions, we can reason about likely future trajectories of the field. For example, as Moore’s law continues to slow down and hardware specialization accelerates [49], emerging computing substrates such as quantum processors and domain-specific accelerators are unlikely to appear uniformly across the periodic space. Instead, consistent with the trends observed in existing system properties, their high cost, operational complexity, and reliability requirements suggest that they will initially be positioned in higher-level computing tiers, most notably in cloud environments [50, 51]. Moreover, to preserve desirable system properties such as democratization and ease-ofuse while integrating these specialized resources, new abstraction 4

layers will be required. These abstractions will shift solutions upward in the periodic space, enabling developers to exploit novel hardware capabilities without deep hardware-specific expertise. As we can see, the periodic space not only describes current solutions, but also can provide a structured lens to predict how future technologies will reshape system behavior across tiers and abstraction levels. To support deeper exploration of these trends for different solutions, we have prepared an interactive webpage0 that allows users to select a property and visualize its behavior across the periodic space for a given solution.

systems must assume failures as part of normal operation, and resilience mechanisms that enable systems to continue functioning seamlessly despite ongoing disruptions become an essential property. However, achieving such resilience becomes increasingly challenging as systems scale across tiers, where complexity grows along multiple dimensions, including heterogeneity, connectivity patterns, and software dependencies. This makes reliability difficult to maintain at low abstraction levels, such as individual devices or infrastructure components. Instead, reliability efforts must shift toward higher abstraction layers, where complexity is hidden behind high-level service guarantees and simplified operational models.

3

Emerging Distributed Systems through the Lens of Periodic Space 3.1 Overview: Agent Abstraction and Sky Tier

3.2.2 Fluidity. The future Sky tier will demand abstractions to enable truly seamless interoperability (aka fluidity), such that computation can migrate as seamlessly as data packets do today [7]. Such fluidity should support both vertical migration, across edgecloud tiers, or horizontal, across multi-cloud. As an example of vertical fluidity consider a group of youth playing game on a coffee shop edge server. Upon arrival of a disabled customer, the game process should seamlessly migrate to cloud to free resources for the disabled user; migration in the opposite direction can happen once the user leaves the place [6]. Horizontal fluidity, however, is getting popular across space datacenters, such as those in low-Earth orbit to maintain services deployed in a certain geographic region [63, 64].

As both infrastructure and applications continue to grow, the periodic space will be expanded in both abstraction and tier dimensions. Abstractions. As resource capacity is continually increasing across all tiers, solutions that were traditionally confined to upper level tiers (e.g., cloud-based FaaS [52]) will be applicable at lower tiers (e.g., edge). However, increasing capability at the lower tiers also brings greater hardware heterogeneity and complexity, demanding for stronger abstractions in those tiers. For instance, Container as a Service (CaaS) [53] and WebAssembly (WASM) [54] abstractions will be used at the “device” tier to democratize complex deployments [55]. To manage the increasing complexity, we envision that the abstraction spectrum will also continue to grow towards: (a) new programming-model abstractions that unify access to diverse specialized hardware across multiple tiers [40, 56]; and (b) higher-level agent abstractions, where developers and end-users can rapidly construct personalized applications on demand across the continuum using LLM-driven agentic systems [57], instead of relying solely on fixed, off-the-shelf SaaS offerings. Agents operate above applications by encapsulating intent, goals, and adaptive behavior [43]. They abstract away both application-specific implementation details [58] and underlying system complexities. Sky Tier. On the tiers dimension, escalating computational demand, particularly from AI, will push the tiers upward—demanding for more interoperability across clouds and hyperscale supercomputing centers towards a “Sky tier” that can collectively meet unprecedented workload volumes. This tier represents a highly federated, globally distributed, and predominantly autonomous computing tier composed of multi-cloud, cross-domain, emerging nonterrestrial (orbital, lunar, and deep-space systems), and quantum computing resources [59] to enable artificial superintelligence [60] and other emerging workloads.

3.2

3.2.3 Sustainability. Despite developments in sustainable computing, future distributed systems will continue to be considerable consumers of computing resources, leading to a significant energy consumption and carbon footprint [65, 66]. While energy consumption and the corresponding carbon emissions will be lower at the local/device level, it will drastically increase as we move along the tiers dimension, peaking at the Sky level. Specifically, the increasing use of specialized accelerators will reflect a shift toward matching workload demands with hardware that delivers higher performance per watt [67]. Near-data and in-memory processing techniques will help to reduce the substantial energy cost associated with moving data between memory, storage, and compute units [68]. Hardware systems will be designed to be more energy efficient, but increased energy efficiency at the hardware level will also lead to increased use of these computing resources and will eventually cause more energy consumption (Jevons’ paradox [69]). Hence, we envision that these advances alone will not be sufficient to keep up with the increased computing demands. More fundamentally, the distributed system infrastructures across the entire periodic space must evolve to become grid-aware and adaptive, coordinating compute, storage, and networking resources in response to the fluctuating capacity of the local power grids and availability of renewable energy [70]. We expect that sustainability will no longer be treated as a secondary objective in distributed systems. It will instead become an essential design principle, shaping decisions at every layer of the distributed computing stack. This requires the ability to measure, quantify, and ultimately reduce the energy consumption and carbon emissions at each of these layers. Hence, a key research opportunity lies in rethinking distributed systems to explicitly handle trade-offs among performance, energy consumption, and carbon emissions.

Trends Towards the Sky Tier

3.2.1 Reliability. As distributed systems evolve toward the Sky tier, their scale approaches planetary levels, where infrastructure spans heterogeneous hardware, multiple administrative domains, and diverse physical environments. Network disruptions, hardware faults, software anomalies, and resource variability become inevitable at this scale, making failures no longer rare or exceptional events but continuous operating conditions [61, 62]. Consequently, Sky-tier

3.2.4 Operational and Infrastructure Costs. Extrapolating observed trends across existing tiers, we anticipate that infrastructure cost per unit of compute (CapEx) and operational expenditure (OpEx)

0 Interactive periodic space: https://hpcclab.github.io/periodic-table

5

will continue to decrease as systems evolves into the Sky infrastructures. At first glance, deploying computing infrastructure in Sky domains, incorporating conventional on-Earth data centers, supercomputers, orbital, lunar, interplanetary, and quantum systems may appear prohibitively expensive. However, we anticipate that, over time, both CapEx and OpEx at these tiers may fall below conventional data centers. Particularly, as the technology of spacebased data centers matures, the transportation and deployment cost of computing infrastructure will decrease [71] (e.g., via reusable rockets) and they can bypass the severe land, power, and water constraints on the earth. Sky computing will further benefit from abundant extra-planetary energy sources, such as solar, which can be utilized without many of the terrestrial constraints [72]. The Sky systems are inherently designed for autonomy and minimal human intervention, significantly reducing maintenance overhead in the long run. Nevertheless, thermal management is likely to remain a fundamental challenge, as cooling in space and on other planetary bodies is constrained by environmental conditions and limited heat dissipation mechanisms. However, while radiative cooling remains a fundamental limitation in space, the vacuum of space allows for passive thermal management [73]. Proponents such as Starcloud and Lonestar suggest that the long-term goal is for space-based, solar-powered data centers to have lower operating costs, potentially 97% lower than on Earth once launch costs drop [74].

3.3

envisage to see more domain-specific LLMs emerge that enable agent behaviors to be characterized (and differentiated) clearly. 3.3.3 Security and Trustworthiness. While higher abstraction improves system reliability through standardization and managed services, there is concern that agent abstraction introduces unprecedented security challenges that may degrade trustworthiness [80]. Unlike traditional applications with deterministic execution paths, agents operate autonomously by planning, reasoning, and executing actions across the device-to-Sky continuum with minimal human oversight. This autonomy can expand the attack surface, as agents interact with external APIs, access sensitive data across organizational boundaries, and execute complex, multi-step workflows that are difficult to audit or constrain by conventional access control policies [81, 82]. Organizations cannot reliably foresee agents’ behavior prior to deployment, and emergent capabilities may introduce vulnerabilities, such as prompt injection attacks [83], data exfiltration via tool misuse [84], or privilege escalation across distributed infrastructure [85]. Ensuring trustworthiness will require new security paradigms, including runtime behavioral monitoring [86], sandboxed execution environments with fine-grained permission models [87, 88], and interpretable decision-making frameworks that expose agent reasoning chains for audit [89, 90]. 3.3.4 Operational and Infrastructure Costs. We anticipate that both the OpEx and CapEx of agents and agentic AI services will be higher than those of application abstraction. Traditional applications rely on multiple abstraction layers in the software stack but are predominantly executed on CPUs and conventional computing resources. On one hand, agentic AI systems require substantial computational resources for both model training and inference, resulting in increased infrastructure and operational costs. On the other hand, such agents are expected to significantly reduce the cost of application development, deployment, and maintenance, which, in the long run, is likely to outweigh the additional resource expenditures.

Trends Towards Agent Abstraction

3.3.1 Democratization. While distributed and cloud computing has permeated nearly every domain of life, its programming has remained exclusive to a small elite of highly specialized organizations [40]. We envision that developments in the abstraction dimension will reverse this trajectory. Computing tiers will evolve into cognitive partners equipped with AI-guided (agentic) programming abstractions [75] that can model device-to-Sky continuum as one uniform compute-base and offer features such as self-composing workflows [57] and declarative deployment [41]. These will collapse the barrier between intent and implementation, and ultimately, make application development as accessible as writing a document.

3.3.5 Governance. Agentic AI systems operate autonomously across distributed and heterogeneous resources, often spanning multiple owners and administrative domains while making real-time decisions that affect users, applications, and infrastructure. These systems commonly adopt a human-on-the-loop model [91], where human intervention occurs only when necessary. As indicated by the periodic space, higher levels of abstraction delegate increasing control and operational responsibility to the underlying infrastructure. Thus, at the Agent abstraction level, governance is essential to ensure accountability, fairness, safety, and compliance despite decentralized control and partial trust among stakeholders. Approaches such as multi-level value alignment [92] aim to keep autonomous decisions aligned with human, organizational, and societal objectives. Addressing these challenges requires scalable monitoring, data-driven insights [93], and adaptive governance mechanisms that integrate policy enforcement, explainable decision-making, and sustainability-aware resource management to enable responsible deployment of agentic AI systems.

3.3.2 Agentic Systems. We envision that Agentic systems will dominate the use of distributed infrastructure across all tiers–from device with a single agent, to other tiers (edge-to-Sky) with multiple agents interacting. The complexity of agent “behaviors” may range from: (i) Well-defined, state-oriented behaviors which are triggered through external events or agent priorities. In this instance, agent actions are constrained and limited by the type and range of events they observe to trigger such actions [76]. (ii) Dynamically updated behaviors which may be defined using domain-specific LLMs [77]. Such behaviors may also be composed dynamically, taking account of capacity and expertise of multiple agents. An example of (ii) is the use of either domain-specific or general purpose LLM models to characterize the agent’s behavior. With agents behaving non-deterministically, understanding their interactions to solve complex/unknown tasks will become a challenge [78]. The multi-agent coordination strategies may involve pre-defined interactions between agents, or may be dynamically developed by discovering the “expertise” of an agent [79], and the use of an LLM (agent) routing mechanism that is able to re-formulate and distribute a request across agents until it is accomplished. We

4

Conclusions

Inspired from the popular chemical periodic table, in this visional paper, we proposed a periodic space based on a continuum of computing tiers and abstraction levels that can capture the behavior 6

of various solutions in the distributed computing landscape. The periodic space also has the ability to anticipate the behavior of emerging solutions in the landscape. We used the periodic space to characterize the noticeable trends in the landscape. Given these trends, we anticipate structural changes will happen to the distributed computing landscape, and when considering the changes under periodic space, we can see strong forces to push innovation toward the high-abstraction, high-tier area.

datacenters on local communities,” in Proceedings of the ACM SIGCAS/SIGCHI Conference on Computing and Sustainable Societies, ser. COMPASS ’25, 2025, p. 769–774. [18] D. V. Black, “Review of The Periodic Table: Its Story and Its Significance,” Technology and Culture, vol. 51, no. 1, pp. 287–289, 2010. [19] L. N. Huynh, Y. Lee, and R. K. Balan, “Deepmon: Mobile gpu-based deep learning framework for continuous vision applications,” in Proceedings of the 15th Annual International Conference on Mobile Systems, Applications, and Services, 2017, pp. 82–95. [20] T. Yeon, V. Xu, H. Hoffmann, and K. Ahuja, “Watchhar: Real-time on-device human activity recognition system for smartwatches,” in Proceedings of the 27th International Conference on Multimodal Interaction, 2025, pp. 387–394. [21] G. Nain, K. Pattanaik, and G. Sharma, “Towards edge computing in intelligent manufacturing: Past, present and future,” Journal of Manufacturing Systems, vol. 62, pp. 588–611, 2022. [22] M. Satyanarayanan, Q. Dong, J. Xu, and P. Pillai, “Towards mobile AI that is accurate and fast,” in Proceedings of the 7th IEEE International Conference on Cognitive Machine Intelligence, ser. COGMI ’25, Nov. 2025. [23] A. Mokhtari, S. Ghafouri, P. Jamshidi, and M. A. Salehi, “HEET: A performance measure to quantify heterogeneity in distributed computing systems,” in Proceedings of the 17th IEEE/ACM International Conference on Utility and Cloud Computing, ser. UCC ’24, 2024, pp. 17–26. [24] B. Charyyev, E. Arslan, and M. H. Gunes, “Latency comparison of cloud datacenters and edge servers,” in Proceedings of the 2020 IEEE Global Communications Conference, ser. GLOBECOM ’20. IEEE, 2020, pp. 1–6. [25] A. Mokhtari, M. A. Hossen, P. Jamshidi, and M. A. Salehi, “FELARE: Fair scheduling of machine learning tasks on heterogeneous edge systems,” in Proceedings of the 15th IEEE International Conference on Cloud Computing, ser. CLOUD ’22, 2022, pp. 459–468. [26] I. Stoica and S. Shenker, “From cloud computing to sky computing,” in Proceedings of the Workshop on Hot Topics in Operating Systems, ser. HotOS ’21, 2021, p. 26–32. [27] J. Liu, Y. Shi, Z. M. Fadlullah, and N. Kato, “Space-air-ground integrated network: A survey,” IEEE Communications Surveys & Tutorials, vol. 20, no. 4, pp. 2714–2741, 2018. [28] B. Denby and B. Lucia, “Orbital edge computing: Nanosatellite constellations as a new class of computer system,” in Proceedings of the 25th International Conference on Architectural Support for Programming Languages and Operating Systems, ser. ASPLOS ’20, 2020, pp. 939–954. [29] Raspberry Pi, “Raspberry Pi,” 2026, accessed: 2026-02-12. [Online]. Available: https://www.raspberrypi.com/ [30] NVIDIA Corporation, “Jetson Nano developer kit,” 2026, accessed: 2026-02-12. [Online]. Available: https://developer.nvidia.com/embedded/jetson-nano [31] Y. Zhang, M. Liu, H. Wang, Y. Ma, G. Huang, and X. Liu, “Research on webassembly runtimes: A survey,” ACM Transactions on Software Engineering and Methodology, vol. 34, no. 8, pp. 1–47, 2025. [32] J. Ménétrey, M. Pasin, P. Felber, and V. Schiavoni, “Webassembly as a common layer for the cloud-edge continuum,” in Proceedings of the 2nd Workshop on Flexible Resource and Application Management on the Edge (FRAME ’22), 2022, p. 3–8. [33] E. Jonas, J. Schleier-Smith, V. Sreekanti, C.-C. Tsai, A. Khandelwal, Q. Pu, V. Shankar, J. Carreira, K. Krauth, N. Yadwadkar et al., “Cloud programming simplified: A Berkeley view on serverless computing,” arXiv preprint arXiv:1902.03383, 2019. [34] K. Morris, Infrastructure as Code: Designing and Delivering Dynamic Systems for the Cloud Age, 3rd ed. O’Reilly Media, 2025. [35] Knative Authors. (2026) Knative official website. [Online]. Available: https: //knative.dev/ [36] Amazon Web Services, Inc. (2026) AWS Step Functions. [Online]. Available: https://aws.amazon.com/step-functions/ [37] H. D. Nguyen and A. A. Chien, “Efficient performance guarantees for function-asa-service with cloud allocators,” in Proceedings of the 26th International Middleware Conference, ser. Middleware ’25, 2025, p. 99–113. [38] H. D. Nguyen, C. Zhang, Z. Xiao, and A. A. Chien, “Real-time serverless: Enabling application performance guarantees,” in Proceedings of the 5th International Workshop on Serverless Computing, 2019, pp. 1–6. [39] P. Lertpongrujikorn, H. D. Nguyen, and M. A. Salehi, “Streamlining cloud-native application development and deployment with robust encapsulation,” in Proceedings of the ACM Symposium on Cloud Computing, ser. SoCC ’24, 2024, p. 847–865. [40] P. Lertpongrujikorn and M. A. Salehi, “Object as a service: Simplifying cloudnative development through serverless object abstraction,” IEEE Transactions on Computers, vol. 75, no. 2, pp. 423–434, 2026. [41] P. Lertpongrujikorn, J. Kwon, H. D. Nguyen, and M. Amini Salehi, “EdgeWeaver: Accelerating IoT application development across edge-cloud continuum,” in Proceedings of the 40th IEEE International Parallel and Distributed Processing Symposium, ser. IPDPS ’26, May. 2026. [42] Microsoft Corporation. (2026) Microsoft office. [Online]. Available: https: //www.office.com/ [43] J. S. Park, J. O’Brien, C. J. Cai, M. R. Morris, P. Liang, and M. S. Bernstein, “Generative agents: Interactive simulacra of human behavior,” in Proceedings of the 36th

Acknowledgments We would like to thank Dr. Mahadev Satyanarayanan, at Carnegie Melon University for his advice and mentoring throughout this study. Also, we appreciate feedback and comments from Dr. Christopher Stewart, at Ohio State University, and Dr. Alexandru Iosup, at Vrije Universiteit Amsterdam. This research is supported by the National Science Foundation (NSF) through CNS CAREER Award# 2419588.

References [1] E. Lagunas, S. Chatzinotas, and B. Ottersten, “Low-earth orbit satellite constellations for global communication network connectivity,” Nature Reviews Electrical Engineering, vol. 1, no. 10, pp. 656–665, 2024. [2] S. Ma, Y. C. Chou, H. Zhao, L. Chen, X. Ma, and J. Liu, “Network characteristics of leo satellite constellations: A starlink-based measurement from end users,” in IEEE INFOCOM 2023-IEEE Conference on Computer Communications. IEEE, 2023, pp. 1–10. [3] M. Armbrust, A. Fox, R. Griffith, A. D. Joseph, R. Katz, A. Konwinski, G. Lee, D. Patterson, A. Rabkin, I. Stoica et al., “A view of cloud computing,” Communications of the ACM, vol. 53, no. 4, pp. 50–58, 2010. [4] L. Atzori, A. Iera, and G. Morabito, “The internet of things: A survey,” Computer networks, vol. 54, no. 15, pp. 2787–2805, 2010. [5] M. Satyanarayanan, W. Gao, and B. Lucia, “The computing landscape of the 21st century,” in Proceedings of the 20th International Workshop on Mobile Computing Systems and Applications, ser. HotMobile ’19, 2019, p. 45–50. [6] T. Chanikaphon and M. A. Salehi, “UMS: Live migration of containerized services across autonomous computing systems,” in Proceedings of the IEEE Global Communications Conference, 2023, pp. 467–472. [7] S. Manatura, T. Chanikaphon, C. Chantrapornchai, and M. Amini Salehi, “FastMig: Leveraging FastFreeze to Establish Robust Service Liquidity in Cloud 2.0,” in Proceedings of the 17th IEEE International Conference on Cloud Computing, ser. CLOUD ’24, 2024, pp. 81–90. [8] N. Akbari, J. Grundy, A. Cheema, and A. N. Toosi, “IntentContinuum: Using LLMs to support intent-based computing across the compute continuum,” in Proceedings of the 2025 IEEE International Conference on Web Services, ser. ICWS ’25, 2025, pp. 573–583. [9] The White House, “National strategy for critical and emerging technologies,” Executive Office of the President, Washington, DC, Tech. Rep., Oct. 2020. [Online]. Available: https://trumpwhitehouse.archives.gov/wp-content/uploads/ 2020/10/National-Strategy-for-CET.pdf [10] European Commission, “Shaping Europe’s digital future,” European Commission, Brussels, Tech. Rep. COM(2020) 67 final, Feb. 2020. [Online]. Available: https://eur-lex.europa.eu/legal-content/en/TXT/?uri=CELEX:52020DC0067 [11] S. Duan, D. Wang, J. Ren, F. Lyu, Y. Zhang, H. Wu, and X. Shen, “Distributed artificial intelligence empowered by end-edge-cloud computing: A survey,” IEEE Communications Surveys & Tutorials, vol. 25, no. 1, pp. 591–624, 2022. [12] R. Bommasani, D. A. Hudson, E. Adeli, R. Altman, S. Arora, S. von Arx, M. S. Bernstein, J. Bohg, A. Bosselut, E. Brunskill et al., “On the opportunities and risks of foundation models,” arXiv preprint arXiv:2108.07258, 2022. [13] T. N. Theis and H.-S. P. Wong, “The end of Moore’s law: A new beginning for information technology,” Computing in Science & Engineering, vol. 19, no. 2, pp. 41–50, 2017. [14] N. C. Thompson and S. Spanuth, “The decline of computers as a general purpose technology,” Communications of the ACM, vol. 64, no. 3, pp. 64–72, 2021. [15] United Nations Conference on Trade and Development, “Digital economy report 2019: Value creation and capture: Implications for developing countries,” United Nations, Geneva, Tech. Rep., 2019. [Online]. Available: https://unctad.org/publication/digital-economy-report-2019 [16] World Bank, World Development Report 2016: Digital Dividends. Washington, DC: World Bank, 2016. [17] W. M. Ngata, N. Bashir, M. Westerlaken, L. Liote, Y. Chandio, and E. Olivetti, “The cloud next door: Investigating the environmental and socioeconomic strain of 7

[68] S. Ghose, A. Boroumand, J. S. Kim, J. Gómez-Luna, and O. Mutlu, “Processing-inmemory: A workload-driven perspective,” IBM Journal of Research and Development, vol. 63, no. 6, pp. 3–1, 2019. [69] B. Alcott, “Jevons’ paradox,” Ecological Economics, vol. 54, no. 1, pp. 9–21, 2005. [70] E. Rodrigues, J. Goldverg, and T. Kosar, “Carbon-aware temporal data transfer scheduling across cloud datacenters,” Proceedings of the IEEE 18th International Conference on Cloud Computing, 2025. [71] S.-Y. Kang, M.-S. Jo, J.-Y. Choi, and S. S. Yang, “Cost effectiveness of reusable launch vehicles depending on the payload capacity,” Aerospace, vol. 12, no. 5, p. 364, 2025. [72] J. Mankins, The Case for Space Solar Power. Virginia Edition Publishing Houston, TX, USA, 2014. [73] Starcloud. (2026) Starcloud official website. [Online]. Available: https://www. starcloud.com/ [74] S. Greengard, “Datacenters go to space,” https://cacm.acm.org/news/datacentersgo-to-space/, 2025. [75] S. Hosseini and H. Seilani, “The role of agentic AI in shaping a smart future: A systematic review,” Array, p. 100399, 2025. [76] M. Colledanchise and P. Ögren, Behavior Trees in Robotics and AI: An Introduction. CRC Press, 2018. [77] M. Li, Y. Zhao, B. Yu, F. Song, H. Li, H. Yu, Z. Li, F. Huang, and Y. Li, “ApiBank: A comprehensive benchmark for tool-augmented LLMs,” arXiv preprint arXiv:2304.08244, 2023. [78] T. Schick, J. Dwivedi-Yu, R. Dessì, R. Raileanu, M. Lomeli, E. Hambro, L. Zettlemoyer, N. Cancedda, and T. Scialom, “Toolformer: Language models can teach themselves to use tools,” Advances in Neural Information Processing Systems, vol. 36, pp. 68 539–68 551, 2023. [79] A. Ghafarollahi and M. J. Buehler, “SciAgents: Automating scientific discovery through bioinspired multi-agent intelligent graph reasoning,” Advanced Materials, vol. 37, no. 22, p. 2413523, 2025. [80] A. G. Chigullapally, S. Vittala, R. F. Hussian, and M. A. Salehi, “Downsides of smartness across edge-cloud continuum in modern industry,” arXiv preprint arXiv:2603.29289, 2026. [81] Z. Deng, Y. Guo, C. Han, W. Ma, J. Xiong, S. Wen, and Y. Xiang, “AI agents under threat: A survey of key security challenges and future pathways,” ACM Computing Surveys, vol. 57, no. 7, pp. 1–36, 2025. [82] F. He, T. Zhu, D. Ye, B. Liu, W. Zhou, and P. S. Yu, “The emerged security and privacy of LLM agent: A survey with case studies,” ACM Computing Surveys, vol. 58, no. 6, pp. 1–36, 2025. [83] OWASP Foundation, “LLM01: 2025 prompt injection,” OWASP Gen AI Security Project, 2025. [Online]. Available: https://genai.owasp.org/llmrisk/llm01-promptinjection [84] Q. Zhan, Z. Liang, Z. Ying, and D. Kang, “InjecAgent: Benchmarking indirect prompt injections in tool-integrated large language model agents,” arXiv preprint arXiv:2403.02691, 2024. [85] Z. Ji, D. Wu, W. Jiang, P. Ma, Z. Li, Y. Gao, S. Wang, and Y. Li, “Taming various privilege escalation in LLM-based agent systems: A mandatory access control framework,” arXiv preprint arXiv:2601.11893, 2026. [86] A. Brown and M. Saner, “The agentic AI security scoping matrix: A framework for securing autonomous AI systems,” AWS Security Blog, November 2025. [Online]. Available: https://aws.amazon.com/it/blogs/security/the-agentic-aisecurity-scoping-matrix-a-framework-for-securing-autonomous-ai-systems/ [87] B. Royal, “Introducing agent sandbox: Strong guardrails for agentic AI on Kubernetes and GKE,” Google Cloud Blog, November 2025. [Online]. Available: https://cloud.google.com/blog/products/containers-kubernetes/agentic-aion-kubernetes-and-gke [88] R. Harang, “Practical security guidance for sandboxing agentic workflows and managing execution risk,” NVIDIA Technical Blog, January 2026. [89] G. Liang, S. Michielssen, and S. Fattahi, “Enhancing performance of explainable AI models with constrained concept refinement,” arXiv preprint arXiv:2502.06775, 2025. [90] A. Jonker, A. Gomstyn, and A. McGrath, “What is AI transparency?” IBM Think Topics, November 2025. [91] J. E. Fischer, C. Greenhalgh, W. Jiang, S. D. Ramchurn, F. Wu, and T. Rodden, “In-the-loop or on-the-loop? Interactional arrangements to support team coordination with a planning agent,” Concurrency and Computation: Practice and Experience, vol. 33, no. 8, p. e4082, 2021. [92] W. Zeng, H. Zhu, C. Qin, H. Wu, Y. Cheng, S. Zhang, X. Jin, Y. Shen, Z. Wang, F. Zhong, and H. Xiong, “Multi-level value alignment in agentic AI systems: Survey and perspectives,” arXiv preprint arXiv:2506.09656, 2025. [93] P. K. Donta, B. Sedlak, V. Casamayor Pujol, and S. Dustdar, “Governance and sustainability of distributed continuum systems: A big data approach,” Journal of Big Data, vol. 10, no. 1, p. 53, 2023.

annual acm symposium on user interface software and technology, 2023, pp. 1–22. [44] Y. Fu, D. Wang, W. Ying, X. Wang, X. Zhang, H. Liu, and J. Pei, “Autonomous data agents: A new opportunity for smart data,” arXiv preprint arXiv:2509.18710, 2025. [45] C. Bird, D. Ford, T. Zimmermann, N. Forsgren, E. Kalliamvakou, T. Lowdermilk, and I. Gazit, “Taking flight with Copilot: Early insights and opportunities of ai-powered pair-programming tools,” Queue, vol. 20, no. 6, pp. 35–57, 2022. [46] X. Li, S. Wang, S. Zeng, Y. Wu, and Y. Yang, “A survey on LLM-based multi-agent systems: Workflow, infrastructure, and challenges,” Vicinagearth, vol. 1, no. 1, p. 9, 2024. [47] M. Satyanarayanan, “The emergence of edge computing,” Computer, vol. 50, no. 1, pp. 30–39, 2017. [48] R. Younis, M. Iqbal, K. Munir, M. A. Javed, M. Haris, and S. Alahmari, “A comprehensive analysis of cloud service models: IaaS, PaaS, and SaaS in the context of emerging technologies and trend,” in Proceedings of the 2024 International Conference on Electrical, Communication and Computer Engineering. IEEE, 2024, pp. 1–6. [49] J. Shalf, “The future of computing beyond Moore’s law,” Philosophical Transactions of the Royal Society A: Mathematical, Physical and Engineering Sciences, vol. 378, no. 2166, p. 20190061, 2020. [50] K. Rallis, I. Liliopoulos, G. D. Varsamis, E. Tsipas, I. G. Karafyllidis, G. C. Sirakoulis, and P. Dimitrakis, “Interfacing quantum computing systems with highperformance computing systems: an overview,” arXiv preprint arXiv:2509.06205, 2025. [51] A. Kumari, R. K. Behera, and B. Sahoo, “Quantum cloud computing: Key technologies, challenges, and opportunities,” Advances in Quantum Inspired Artificial Intelligence: Techniques and Applications, pp. 99–124, 2025. [52] S. Kounev, N. Herbst, C. L. Abad, A. Iosup, I. Foster, P. Shenoy, O. Rana, and A. A. Chien, “Serverless computing: What it is, and what it is not?” Communications of the ACM, vol. 66, no. 9, p. 80–92, 2023. [53] B. Burns, B. Grant, D. Oppenheimer, E. Brewer, and J. Wilkes, “Borg, Omega, and Kubernetes,” Communications of the ACM, vol. 59, no. 5, pp. 50–57, 2016. [54] A. Haas, A. Rossberg, D. L. Schuff, B. L. Titzer, M. Holman, D. Gohman, L. Wagner, A. Zakai, and J. Bastien, “Bringing the web up to speed with WebAssembly,” in Proceedings of the 38th ACM SIGPLAN Conference on Programming Language Design and Implementation, ser. PLDI ’17, 2017, p. 185–200. [55] S. Kakati and M. Brorsson, “A cross-architecture evaluation of WebAssembly in the cloud-edge continuum,” in Proceedings of the IEEE 24th International Symposium on Cluster, Cloud and Internet Computing, ser. CCGrid ’24, 2024, pp. 337–346. [56] P. Lertpongrujikorn and M. A. Salehi, “Object as a service (OaaS): Enabling object abstraction in serverless clouds,” in Proceedings of the 16th IEEE International Conference on Cloud Computing, ser. CLOUD ’23, 2023, pp. 238–248. [57] A. Esashi, P. Lertpongrujikorn, S. Kato, and M. A. Salehi, “Action engine: Automatic workflow generation in FaaS,” Future Generation Computer Systems, vol. 174, p. 107947, 2026. [58] A. Esashi, P. Lertpongrujikorn, J. Makino, Y. Fujimoto, and M. Amini Salehi, “Foundation CAN LM: A pretrained language model for automotive CAN data,” in Proceedings of the 37th IEEE Intelligent Vehicles Symposium, Detroit, MI, USA, Jun. 2026. [59] P. V. Trinh and S. Sugiura, “Quantum Internet in the sky: Vision, challenges, solutions, and future directions,” IEEE Communications Magazine, vol. 62, no. 10, pp. 62–68, 2024. [60] A. Grabowska and A. Gunia, “On quantum computing for artificial superintelligence,” European Journal for Philosophy of Science, vol. 14, no. 2, p. 25, 2024. [61] B. Nicolae, J. M. Wozniak, T. Bicer, H. Nguyen, P. Patel, H. Pan, A. Gueroudji, M. Gonthier, V. Hayot-Sasson, E. Huerta et al., “Diaspora: Resilience-enabling services for real-time distributed workflows,” in 2024 IEEE 20th International Conference on e-Science, ser. e-Science ’24, 2024, pp. 1–9. [62] H. D. Nguyen, T. Bicer, B. Nicolae, R. Kettimuthu, E. A. Huerta, and I. T. Foster, “Resilient execution of distributed x-ray image analysis workflows,” Frontiers in High Performance Computing, vol. 3, p. 1550855, 2025. [63] T. Pfandzelter and D. Bermbach, “Komet: A serverless platform for low-earth orbit edge services,” in Proceedings of the 2024 ACM Symposium on Cloud Computing, ser. SoCC ’24, 2024, p. 866–882. [64] V. Bhosale, A. Gavrilovska, and K. Bhardwaj, “Krios: Scheduling abstractions and mechanisms for enabling a LEO compute cloud,” in Proceedings of the ACM Symposium on Cloud Computing, ser. SoCC ’24, 2024, p. 322–340. [65] A. Imran, T. Kosar, J. Zola, and M. F. Bulut, “Towards sustainable cloud software systems through energy-aware code smell refactoring,” in Proceedings of the IEEE 17th International Conference on Cloud Computing, ser. CLOUD ’24. IEEE, 2024, pp. 223–234. [66] J. Goldverg, H. Jamil, E. Rodrigues, and T. Kosar, “Towards carbon-aware data transfers,” IEEE Internet Computing, vol. 29, no. 2, pp. 19–26, 2025. [67] J. L. Hennessy and D. A. Patterson, “A new golden age for computer architecture,” Communications of the ACM, vol. 62, no. 2, pp. 48–60, 2019.

8