Demo and Poster Night Information
Location
AC Hotel Washington DC Downtown
PH Level
1112 19th Street NW, Washington, District of Columbia, USA, 20036
Date & Time
September 29, 2026
5:30-7:30 pm ET
Demo night presenters: Please arrive to set up your demo no later than 5:15 pm. The event will begin promptly at 5:30 pm
Guests: Please relax in the lobby until 5:30 pm to allow presenters to set up their demos. Thank you!
Demo and Poster Night Participants
Presenter: Manas Das
This showcase introduces an extensible AI cluster resource monitoring tool that delivers live, browser-based performance dashboards for Fabric users. This tool can be extended to other platforms/testbeds. It provides comprehensive visibility from the kernel to the application level, complete with support for custom monitoring code integration.
Presenter: Nik Sultana
This demo showcases a new approach that provides sound, machine-checked guarantees over the analysis of network data. This approach was developed to ease the sharing of network data for research, and it builds on the Patchwork system that runs on FABRIC.
Presenter: Venkat Sai Suman Lamba Karanam
Data-intensive workflows in high-throughput computing (HTC) move large amounts of data across sites connected via a dedicated distributed WAN. Timely delivery of required data (for data locality) not only has direct impacts on the end-end job performance but also on the resource utilization (bandwidth, compute). We developed and deployed the core adaptive compression methodology on Chameleon, FABRIC, and other testbeds for validation with full layer-2 control (happy to share results of this validation). This work is aimed at reducing the data transfer costs in distributed WAN ecosystems; particular domains include High-Energy Physics, large-scale genomics and climate modeling. Our current extension includes Rust and Lean based compression implementations of existing compression algorithms (lz4, bz2, snappy, lzma, zstd & sz, zfp, mgard). We found that the combination of implementation+compression algorithm choice is just as important. For example, we saw combinatorial gains in compress/decompress overheads. We will also discuss our prototypes Julia implementation of these compression algorithms and the consequential trade-offs.
Presenter: Qiang Liu and Ming Zhao
In this poster, we propose AdaptiveOn, a new wireless networking framework, to support deterministic round-trip networking for network-initiated traffic in 5G. The key idea is to proactively allocate the uplink grants to UEs with sufficient transport block size (TBS), so that UEs can send turnaround traffic immediately, without going through the conventional uplink request-report-grant pipeline. By designing four integral modules (payload estimator, timing learner, MCS adapter, and temporal averager), AdaptiveOn substantially reduces the queuing, segmentation, and re-transmission delay in both uplink and downlink transmission.
Presenter: Alexander Wolosewicz
Historical experience in the High-Performance Computing community teaches us that as computing systems grow, instances of failures go from rare to regular occurrences. A survey of the growth in the size and complexity of Data AcQuisition (DAQ) networks in High-Energy Physics (HEP) experiments reveals that these networks are scaling exponentially, trending to a point where automated fault handling should be considered over the current manual practice. We propose a general system, DiDAQt, which is designed to provide fault detection and handling in HEP DAQs specifically, through MPI-like primitives that allow it to be added easily to existing systems. We evaluate the scalability and response time of a prototype on the FABRIC national testbed, with one of the largest experimental setups run on it, with results indicating sufficient scalability for current and near-future DAQs as well as practical response times (under 1 microsecond decision time).
Presenter: Sravani Kurma and Charles Ueltschey
This demo presents a mission-aware agentic O-RAN framework for UAV communications that combines mobility-dependent wireless context from the AERPAW Digital Twin with programmable resource control using the OAIC testbed. A local large language model (LLM) reasons over mission priority, traffic demand, link conditions, and network KPIs to generate constrained resource-allocation decisions that are applied and verified through NexRAN. The demonstration illustrates how critical command, control, and telemetry traffic can be dynamically prioritized as UAV communication conditions and mission requirements change.
Presenter: Kevin Hermstein
We demonstrate the mmWave and edge-computing capabilities of the COSMOS testbed. We utilize a mobile 28 GHz phased array antenna module and a Velodyne LiDAR to emulate the sensor(s) on an autonomous vehicle (AV). The AV transmits the LiDAR data to a COSMOS node where the LiDAR data is then processed at the edge-cloud for smart city applications. Thanks to the high speed mmWave connection and powerful compute processing available in COSMOS, all of this is achieved in a real-time, practical system, demonstrating real world feasibility for edge-cloud based processing of sensor data for real-time control.
Presenter: Osamah Alzacko
The Open RAN (O-RAN) E2 interface is a security-critical control channel that lacks mandatory transport security, leaving both the Near-RT RIC and E2 Nodes vulnerable to complex stateful message manipulation. We demonstrate EeTwo, a state-aware, gray-box fuzzer that comprehensively tests both endpoints through dual-role peer impersonation and dual-layer ASN.1 type-confusion mutation. This demo showcases how EeTwo synthesizes compliant E2AP/E2SM traffic to navigate strict protocol states and expose deeply nested memory-safety vulnerabilities, highlighting pre-authentication denial-of-service faults discovered across major production O-RAN stacks.
Presenter: Charles Ueltschey and Dr. Vuk Marojevic
We will showcase RAN Tester UE and its ability to run automated tests against 5G random access networks. These tests include SSB spoofing, RACH flooding, uplink CP DoS, synchronization attacks, sniffing and more. All will be shown being run on real hardware with agentic AI MCP configuring the attacks.
Presenter: Chandrima Ghatak
We present SPHERE-CAN, a platform that exposes a physical automotive network testbed through the SPHERE research infrastructure. The system connects hardware at Colorado State University with the USC Information Sciences Institute SPHERE research infrastructure, allowing users to configure experiments, interact with embedded controllers, and observe network traffic. This demo shows monitoring, message injection, and attack execution on live systems, along with logging of resulting behavior. By enabling interaction with real automotive networks, SPHERE-CAN supports reproducible experimentation and evaluation of automotive cybersecurity.
Presenter: Taimoor UI and Joshua Ofori Boateng
This demo runs live on ARA, the NSF PAWR wireless living lab spanning 500+ square miles of rural Iowa, where an open-source OCUDU 5G gNB serves a Linux Foundation Duranta UE on a scouting drone that streams crop imagery to an edge AI weed detector for precision spot spraying. To ensure predictable, real-time performance of the wireless links that the physical AI depends on, we implement KaiAir in OCUDU as an open source contribution, bringing our PktR interference and power control and LDP real-time scheduling algorithms into a open-source 5G stack. A live dashboard lets the audience turn KaiAir's knobs and watch SINR, latency, throughput, and detection performance. Resources shown in this demo are open to the research community for their own at-scale experiments on Physical AI for real-world rural applications.
Presenter: Zhangyu Guan
We will demonstrate UnionLabs, a cloud-based federated experimentation platform that integrates heterogeneous wireless testbeds across multiple institutions and enables users to schedule and conduct real-time over-the-air (OTA) experiments remotely. The demo will showcase end-to-end experiment workflows across federated testbeds. Through UnionLabs, we aim to democratize access to diverse wireless research infrastructure and foster a community-driven ecosystem for reproducible experimentation.
Presenter: Rohan Bail
This demo showcases information-theoretically covert communication over a live RF channel using COSMOS testbed. Using a pre-shared secret key and sparse signaling, Alice communicates reliably with Bob while preventing adversary Willie from reliably determining whether Alice is transmitting or silent. Attendees can provide input at the transmitter, view the recovered message at the receiver, and monitor Willie’s real-time observations and detection statistics. The experiment demonstrates how readily available COSMOS resources can be combined to remotely implement custom low-level wireless protocols, create controlled AWGN conditions, and build a compute cluster for real-time processing.
Presenter: Samuel Koontz
EduceLab is a mid-scale heritage science facility serving a diverse community of scholars, curators, conservators, archaeologists, and other researchers working with complex cultural heritage materials. This presentation will discuss how the facility’s infrastructure enables multimodal data acquisition and analysis through tools including spectral and photogrammetry systems, optical 3D scanning, and Raman spectroscopy, providing a more holistic understanding of heritage objects while minimizing physical interaction with these sensitive materials. Case studies, including painted objects and carbonized papyrus from Petra, will demonstrate how complementary techniques can be combined to address research questions that would be difficult or impossible to resolve using a single analytical approach. The demonstration will also highlight how this flexible infrastructure supports a wide range of users, materials, and research needs.
Presenter: Faik Kerem Ors
Security vulnerabilities in O-RAN specifications and open-source implementations require experimental validation to determine whether they can be reproduced in realistic deployments and lead to practical security consequences. We demonstrate an end-to-end framework for validating O-RAN security vulnerabilities using the POWDER over-the-air testbed. Our experiments integrate open-source O-RAN components with POWDER's wireless infrastructure to exercise vulnerable behaviors in a running network and observe their effects across the O-RAN system. The demonstration highlights how programmable wireless testbeds can bridge the gap between vulnerability analysis and real-world experimentation, providing a reproducible environment for evaluating the practical security and interoperability implications of vulnerabilities in emerging O-RAN systems.
Presenter: Amith Gorthi Sriniavasa Prabhakara Narasimha
This work presents a new lab series that teaches DPA (Data Path Accelerator) programming on NVIDIA BlueField-3 DPUs. Users progress through a sequence of hands-on labs, from DPA thread initialization and inter-thread data sharing, through asynchronous host-DPA RPC communication and on-path packet modification, to live telemetry, rate-limited traffic control, and a capstone static firewall with host-pushed rule updates using an atomic valid-flag commit pattern for SRAM consistency. Each lab builds on the DPACC toolchain, DOCA Flow, and the DPA execution-unit (EU) threading model, giving users direct, hands-on experience with the architectural concepts underlying accelerated infrastructure workloads. Together, the labs provide a structured, technically grounded introduction to DPA programming for the next generation of SmartNIC/DPU developers.
Presenter: Akhil Gorthi Bala Sai
This work presents a new lab series that teaches DPA (Data Path Accelerator) programming on NVIDIA BlueField-3 DPUs. Users progress through a sequence of hands-on labs, from DPA thread initialization and inter-thread data sharing, through asynchronous host-DPA RPC communication and on-path packet modification, to live telemetry, rate-limited traffic control, and a capstone static firewall with host-pushed rule updates using an atomic valid-flag commit pattern for SRAM consistency. Each lab builds on the DPACC toolchain, DOCA Flow, and the DPA execution-unit (EU) threading model, giving users direct, hands-on experience with the architectural concepts underlying accelerated infrastructure workloads. Together, the labs provide a structured, technically grounded introduction to DPA programming for the next generation of SmartNIC/DPU developers."Hands-on networking education relies on laboratories where learners configure networked systems, observe behavior, and debug failures. Virtual labs make this experience scalable, but support remains difficult because useful guidance often requires knowing both the intended procedure and the current state of a learner's environment. A lab manual describes the topology, setup steps, commands, and expected outputs, but it cannot inspect whether a learner has missed a step, configured the wrong interface, or produced an inconsistent routing state. General-purpose Large Language Models (LLMs) can explain networking concepts, but they are not grounded in a specific lab manual or in the live state of a running lab instance. They may also provide final commands too early, reducing the opportunity for learners to reason through the debugging process.
Presenter: John Drogo and Shivalee Shah
We present the construction of one of the largest United States based quantum internet testbeds. Through collaborations with industry, academic, and national laboratory partners we are expanding the NSF PAWR COSMOS testbed to enable quantum information exchange. We are leveraging dark optical fibers to form a multi-node photon testbed for quantum algorithm and hardware development. We show initial results including characterization of these quantum channels and discussion of unique control and timing challenges encountered building quantum networks.
Presenter: Hakan Gulec
This presentation will share practical experiences running a hands-on MLOps course at NYU, where students build complete machine learning pipelines, from distributed training and experiment tracking to deployment, automation, and monitoring, using Chameleon.
Presenter: Gaurav Kuwar
Hardware acceleration of privacy-preserving computation like Zero-Knowledge Proofs and Fully Homomorphic Encryption demands complex, compute-intensive and very large advanced arithmetic kernels. Currently, research groups largely operate in silos, creating high barriers to entry and fragmented tooling across the community. We present an open-source research infrastructure designed to streamline the exploration, optimization, and evaluation of these kernels. Our framework builds on Catapult high-level synthesis; its automated pipeline integrates logic synthesis (Synopsys Design Compiler), power analysis (PrimePower), and RTL and Gate-Level verification (QuestaSim) for ASICs and FPGA synthesis (Vivado). The framework abstracts away the intricate toolchain configurations and infrastructure adaptations required to make commercial tools effectively handle these large, complex cryptographic kernels. By unifying these flows into an open platform, Cryptolets aims to lower the barrier to entry and provide a common foundation for collaborative research.
Presenter: Saad Elbeleidy
Quori is a low-cost, modular, open-source social robot platform designed with the Human-Robot Interaction (HRI) community for the HRI community. This demo video demonstrates how one can configure Quori in different formats by relying on Quori's various modules and Quori's open modular connector.
Presenter: Shayan Nazeer
This demo presents CellKV, a framework to enable continuous LLM inference as mobile users hand over between cellular edge base stations. Instead of transferring the entire multi-gigabyte KV cache, CellKV moves a bounded, semantically selected working set, streams newly generated KV state during handover, and retrieves or reconstructs omitted context on demand. We demonstrate that CellKV reduces handover-induced generation stalls while preserving access to session context across repeated mobility.
Presenter: Komal Thareja
"We demonstrate a real high-energy physics training workload that automatically migrates across the FABRIC testbed in response to cleaner energy availability, using unmodified Pegasus and HTCondor. Training is divided into checkpoint-linked workflow segments. FABRIC GPU workers publish real-time grid carbon intensity and measured GPU power through HTCondor ClassAds, allowing standard matchmaking to place each segment using a one-line carbon-aware ranking expression. All data transfers use HTCondor file I/O, enabling the data cost of each migration to be measured, while segment-level energy consumption is calculated from worker-side NVML measurements. A live dashboard displays placement decisions, site-specific carbon intensity, and measured CO₂ emissions for each segment. The workflow also generates a report showing training accuracy improving over time, with color changes indicating carbon-driven migrations."