Abstract
Commercial cloud service providers like Amazon Web Services (AWS) and Microsoft Azure impose prohibitive costs and vendor lock-in risks, while public cloud multi-tenancy creates security vulnerabilities for classified defense workflows requiring strict data sovereignty. To provide an open-source alternative capable of meeting these strict security and cost-efficiency goals, this project details the design and deployment of a fully repatriated, on-premises private cloud. We deployed a three-node high-availability cluster on existing bare-metal hardware, leveraging Incus for system container orchestration and Open Virtual Network (OVN) for network virtualization. Unlike standard application containers, Incus system containers simulate full operating systems, while OVN overlay networking enables dynamic multi-tenant provisioning and strict network segmentation through software-defined overlays without physical infrastructure dependencies. This architecture replicates the isolated, scalable topologies of public clouds, offering commercial-grade features including load balancing, auto-scaling, and the automated provisioning of Virtual Private Cloud (VPC) environments. The platform achieved 99% uptime during operational deployment and demonstrated enterprise-grade production readiness. Instance migrations completed within 30 seconds, HAProxy provided sub-second failover, and OVN successfully enforced network isolation between overlapping IP spaces. Ultimately, this project proves that organizations can build robust, customizable infrastructure on-premises that matches commercial service providers in functionality while maintaining absolute data control and eliminating recurring operational costs.
In March 2020, Folding@home surpassed 2.4 exaFLOPS of processing power, mobilizing over one million volunteer devices to simulate COVID-19 protein dynamics. While media outlets praised this rapid mobilization as a "citizen science" revolution, this celebratory framing obscures critical questions of structural power regarding who governs the decentralized infrastructure and captures its generated intellectual property. This research maps the sociotechnical network of Folding@home to evaluate how its technical architecture, organizational governance, and discursive framing shape the distribution of agency and value. Employing Actor-Network Theory (ANT), discourse analysis, and economic analysis, the study investigates the material distribution of control, the divergence between democratic rhetoric and structural reality, and institutional intellectual property capture. The analysis reveals that the platform operates on an asymmetrical infrastructure where volunteers act as essential resource providers but are structurally excluded from intellectual membership or governance. Furthermore, the rhetoric of "citizen science" masks the unilateral extraction of commercial and intellectual value. The economic value generated by uncompensated volunteer labor translates directly into academic publications, institutional prestige, and patentable discoveries captured entirely by university and corporate actors, while volunteers receive only symbolic gamified recognition. Ultimately, Folding@home's rigid centralization reflects discretionary governance choices designed to maintain institutional control rather than technical constraints. As crowd computing expands into highly commercialized domains like artificial intelligence training, these findings shed light on the ideological framing of distributed data harvesting.
Both the STS research and the technical capstone project examine computational infrastructure and control through different lenses. The technical project answers the question of infrastructure control through engineering, while the STS project answers it through analysis. The bare-metal private cloud architecture serves as a direct, material foil to the Folding@home platform. By granting the administrator absolute agency over data routing, node priority, and network segmentation through Open Virtual Network, the technical project demonstrates that infrastructure can be robustly decentralized while retaining local authority. Conversely, the Folding@home client systematically strips the individual node of routing agency, centralizing all allocation authority at the institutional server level. Building a functional decentralized routing architecture proves that the cost and dependency associated with commercial cloud services are market realities that strategic engineering can bypass, and that Folding@home's centralization is a specific governance choice rather than a universal necessity of distributed computing.
