SDNator is Not Another SDN Controller: Enabling Extensible Data-Driven Control in Cyber-Physical Systems
Yikai Lin, Ruowang Zhang, Efe Balta, Jianbin Zhang, Xiao Zhu, Kira Barton, Dawn Tilbury, Z. Morley Mao · 2026
Read SDNator is Not Another SDN Controller: Enabling Extensible Data-Driven Control in Cyber-Physical Systems by Yikai Lin, Ruowang Zhang, Efe Balta, Jianbin Zhang, Xiao Zhu, Kira Barton, Dawn Tilbury, Z. Morley Mao (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2605-23816.txt.
Introduction
Lin and colleagues build and open-source SDNator, the first framework for extensible control in cyber-physical systems. SDNator uses an application-driven design where apps both use and supply data shaping workflows. Case studies in manufacturing and networking show what it can do and how easy it is to use. SDNator builds the first digital-twin controller for manufacturing fleets. Simulations show SDNator-based scheduling shortens production time and improves reliability amid faults, and adjusts plans for urgent requests. Benchmarks show SDNator scales well and matches Ryu on performance.
Catalog record
| CREATOR | Yikai Lin, Ruowang Zhang, Efe Balta, Jianbin Zhang, Xiao Zhu, Kira Barton, Dawn Tilbury, Z. Morley Mao |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 56053 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2605.23816
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
SDNator is Not Another SDN Controller: Enabling Extensible Data-Driven Control in Cyber-Physical Systems Yikai Lin, Ruowang Zhang, Efe Balta, Jianbin Zhang, Xiao Zhu, Kira Barton, Dawn Tilbury, Z. Morley Mao Abstract An SDN-like centralized control architecture is increasingly popular and has been widely explored in cyber-physical systems (CPS) such as manufacturing, internet-of-things, and autonomous vehicle systems for higher flexibility, programmability and scalability. However, no existing frameworks can offer domain-agnostic, easily extensible support for data-driven CPS applications. In this work, we design, implement, and open-source SDNator , the first framework to enable extensible, data-driven control in CPS. SDNator embraces an application- and data-driven design where applications function as data consumers and producers to collectively define the workflows of the controller. SDNator also incorporates two data store backends to support both event-driven and data-driven programming patterns. Benchmarks show that SDNator is highly scalable, and delivers comparable performance to Ryu, a widely used SDN controller. Moreover, we demonstrate the capabilities and usability of SDNator through our case studies of manufacturing and networking systems. By integrating applications from respective domains, we build different “controllers” for different scenarios. Most notably, we leverage SDNator to implement the first digital-twin-equipped central controller for additive manufacturing fleets. We show through extensive and realistic simulations that SDNator-based scheduling can (1) significantly shorten production time and improve reliability in the presence of anomalies compared to decentralized approaches, and (2) flexibly adjust and optimize production plans upon urgent requests such as producing Personal Protective Equipment during the COVID-19 pandemic. 1 Introduction Recent advances in hardware capabilities ( e.g., programmable switches, autonomous automobile, smart manufacturing) and network connectivity technologies ( e.g., 4G and 5G) have enabled new software-hardware interactions in cyber-physical systems (CPS). For instance, a centralized control architecture inspired by Software-Defined Networks (SDN) has been widely explored as a viable alternative to traditional distributed solutions in various CPS such as manufacturing [ 43 , 50 , 30 ] , internet-of-things (IoT) [ 41 ] , and autonomous driving [ 36 , 42 ] . Applications can utilize aggregated real-time information of a CPS to improve its flexibility, efficiency and reliability. Furthermore, this paradigm opens up exciting research opportunities for new control strategies and workflows that were previously infeasible. However, developing and deploying CPS applications based on this centralized paradigm is non-trivial. Due to a lack of available “controller” frameworks that readily support CPS applications, existing works often resort to extension or adaptation of legacy SDN controllers to tailor to their particular application requirements [ 43 , 41 , 37 ] . Such extension or adaptation is time-consuming, non-reusable, and inherits the same limitation in supporting data-driven CPS applications ( i.e., inefficient data production/consumption, lack of data heterogeneity and historical data support, etc. [ 40 ] ). If a new “controller” variant has to be built every time new CPS applications emerge, which is likely given how heterogeneous CPS are, the value of flexibility and programmability diminishes. To fundamentally address this problem, we design, implement and open-source SDNator (§ 3 ), a framework that enables researchers and developers to easily implement new or incorporate existing CPS applications in a centralized workflow. SDNator is not a specialized controller; rather, it enables building different controllers through plug-and-play of different applications/systems. SDNator embraces an application- and data-driven design where applications function as data consumers and producers to collectively define the workflows and capabilities of a controller. A data-driven design also allows SDNator to be domain- and protocol-independent, making it possible to integrate with and be deployed in real-world CPS. Enlightened by previous work on CPS application characteristics and requirements [ 40 ] , SDNator incorporates two different data backends to provide both event-driven and data-driven programming patterns. One of them serves as persistent storage to enable historical data usage for offline analysis. To make SDNator more instrumental and efficient, we implement an on-demand data production (§ 3.7 ) mechanism to give consumer apps control over what data will be produced at what frequency. Fault tolerance and recovery features (§ 3.8 ) are also built into SDNator for extra reliability. Through extensive benchmarks (§ 5 ), we show that SDNator (1) delivers over 100K msgs/s using one CPU core on a commodity PC, (2) incurs as little as sub-100us end-to-end delay, and (3) scales with 60 geo-distributed apps and high network latency. On top of that, we leverage SDNator to implement 3 different control workflows for manufacturing (§ 6.1 ) and networking (§ 6.2 ) CPS in our case studies. In particular, in light of the “citizen-supply-chain” phenomenon [ 32 ] during the COVID-19 pandemic 1 1 1 Universities, tech firms and 3D print enthusiasts with their own 3D printers respond to the shortage of healthcare workers’ personal protective equipment (PPE) by producing it themselves. , we demonstrate (§ 6.1.4 ) how an SDNator-based centralized scheduling workflow can substantially speed up PPE production by 2X without compromising existing production jobs compared to a decentralized uncoordinated approach. We make the following contributions in the paper: • We design, implement and open-source 2 2 2 https://github.com/anonymous/anonymous.git SDNator, an extensible, data-driven framework for enabling centralized control in cyber-physical systems. SDNator achieves extensibility through plug-and-play of apps with no controller adaptations or modifications. SDNator achieves generality through its data-driven abstractions, allowing it to easily integrate with different CPS. • We design and implement a lightweight client-side library called Data Ubiquity Engine (DUE) that (1) provides well-defined APIs to support both event-driven and data-driven programming patterns, (2) transparently enforces on-demand data production, (3) intelligently performs batching and buffering, and (4) silently monitors application health. DUE enables easy onboarding of heterogeneous apps in SDNator, delivers over 100K msgs/s and incurs less than 100us delay. • We demonstrate how easy it is to enable centralized control using SDNator (either through developing new CPS applications or integrating existing ones) in our case studies of manufacturing and networking CPS. In particular, we carry out the first study on digital-twin-equipped centralized control of additive manufacturing fleets and show that it shortens normal production time by up to nearly 40%, and PPE production time by more than 50% compared to a decentralized baseline approach. 2 Background 2.1 Software-Defined Networks The Software-Defined Networks (SDN) paradigm, in essence, describes a centralized control architecture where applications (the S in SDN) possess the intelligence of the system and fulfill many roles such as computing, decision making, and reconfiguration (of network devices) while leveraging the global view provided by a (logically) centralized controller. Compared to traditional distributed approaches, centralized solutions substantially improve the flexibility and efficiency of device management, simplify and speed up software development and iteration, and open up exciting opportunities for network management thanks to global visibility. Legacy SDN controllers are not for CPS apps. A variety of SDN controllers are available in different programming languages, such as Ryu [ 26 ] , Floodlight [ 7 ] , OpenDaylight [ 45 ] and ONOS [ 12 ] . Legacy SDN controllers are designed as monolithic ecosystems where internal applications are mostly written in the same programming language and run alongside each other ( i.e., scaling is achieved through multiplying controller instances rather than apps). Protocols ( e.g., OpenFlow [ 44 ] ), device abstractions ( e.g., switches) and data types