Experimental systems research is exciting, with multiple interesting directions to explore. One is about building a testbed that mimics the real world so that we can test how a new idea would work in the wild. Another is about building a testbed that is different from the real world, trying to figure out if it would be a better system. For me, programmability played an essential role in both cases. In this talk, I will present two testbeds we have built, each focusing on these two directions. First, I will present P4Campus, a testbed that helps researchers run their network idea in their own campus network against real production traffic. Next, I will present Pronto, an open source end-to-end 5G connected edge with deep programmability. I will share our experiences, lessons learned, and ideas for future work.

This panel considers topics in the life-cycle of a network testbed, everything from deploying one to sustaining it beyond its funding period. This includes, but is not limited to, topics such as federation, measurement and monitoring, data archiving, curation and management, privacy, reproducibility, and technical support, not to mention attracting users and building community(ies) for a testbed. Each panelist has extensive experience and expertise with network testbeds -- beyond those testbeds listed -- and bring their lessons learned, best practices, and ideas to the panel as we look forward to requirements and considerations for future testbeds.

The AERPAW project is an ambitious project, funded by the PAWR program of the US NSF, to created a remote accessible research platform for a research facility with some distinct features that makes its usage model unique, and non-obvious to many researchers desirous of making use of this platform. AERPAW is primarily a physical resource (not a computing or cyber-resource) - the RF enviroment, and the airspace. Experimenters can explore them through radio transceivers and Unmanned Aerial Vehicles, both under the Experimenter's programmatic control. Since the entire workflow of the user is through the mediation of virtual computing environments, users often tend to think of AERPAW as a computing resource, and find some of the experiment workflow counter-intuitive. In this paper, we articulate the challenges and considerations of designing an experiment workflow that balances the need for guaranteeing safe testbed operation, and providing flexible programmatic access to this unique resource.

The Open Cloud Testbed (OCT) provides nodes with Field Programmable Gate Arrays (FPGAs) that are under the complete control of the user and are directly attached to a network switch via two 100Gbps connections. We provide TCP and UDP stacks on the FPGAs. In addition, users have the ability to experiment with their own protocol. We present several experiments which make use of this capability including TCP throughput measurements, an encryption/decryption example, and machine learning inference split across two FPGAs where the images are input on one node and the labelled output available on a second node. The testbed is available for researchers to perform their own experiments, and includes a development platform that allows users to create FPGA applications. Network measurement results show we achieve close to peak bandwidth by tuning appropriate parameters.

The Cooperative Intelligent Transportation System (C-ITS) testbed or simplified called the Smart Highway (Antwerp, Belgium) is designed to facilitate research in the area of distributed/edge computing and vehicular communications. The Smart Highway testbed deploys the Cooperative Awareness Basic Service to exchange Cooperative Awareness Messages (CAMs) between road C-ITS entities, e.g., C-ITS vehicles and Road-Side Units (RSUs). CAMs support vehicular safety and traffic efficiency applications by providing them with the continuous status information of relevant C-ITS entities. Therefore, it is important that those message are delivered with low latency, especially the CAMs that originate from special vehicles, e.g., emergency vehicles, police cars, and fire trucks. In this paper, we research the impact of CAM messages configuration on the communication latency among vehicles. Moreover, we have performed the practical experimentation to evaluate the aforementioned impact, using ITS-G5 and LTE-V2X system under realistic vehicular conditions, on the Smart Highway testbed locate in Antwerp.

We address the design of an acoustic communications testbed with the possibility to implement the acoustic feedback. The design focuses on providing an experimental verification for signal processing algorithms that are applicable to both in-air and underwater scenarios. Feedback highlights the possibilities of acquiring channel conditions, and enables adaptation on the transmit side. We demonstrate our testbed through multi-user multi-carrier uplink and downlink communications and adaptive modulations. The fully-software defined, flexible and expandable testbed provides an ideal platform for the development of the next-generation of acoustic communications. Experimental results show that with the proper feedback, the proposed communication schemes provide excellent performance that without feedback would not have been possible, and making them suitable for practical implementations.

Wide network softwarization is creating fertile ground for the application of novel concepts in the management of the deployed network functions. This allows a drastic shift for the applications hosted on top of the network, as instead of configuring their behaviour to match the network status (network-aware applications) the technology can shift to a self-organizing network that adapts to the hosted applications (service-aware network). In this work, we design, develop and evaluate such a service-aware approach for the telecommunications network. By employing Machine Learning, we are able to classify and predict traffic exchanges made over the network, and appropriately dynamically allocate the slices in the network in real-time. We use as a reference platform the OpenAirInterface framework, and the FlexRAN controller for programming the slice decisions at the RAN level, and evaluate our scheme under real-world settings in a testbed environment.

Time-Sensitive Networking (TSN) is vital to enable time-critical deterministic communication, especially for applications with industrial-grade requirements. IEEE TSN standards are key enablers to provide deterministic and reliable operation of Ethernet networks. However, much of the research is still done in simulated environments or using commercial TSN switches lacking flexibility in terms of hardware and software support. In this work, we evaluate two different Cloud testbeds for TSN experimentation, analyzing their hardware features, the influence of the testbed management infrastructure, and the data plane performance. Furthermore, we present a prototype of a modular Centralized Network Configuration (CNC) controller that facilitates the deployment of Linux-based TSN networks. We identify and discuss the main modules of our controller and evaluate its feasibility by using it to deploy TSN networks on different testbeds. Finally, we provide insights for researchers interested in experimenting with TSN features on open Cloud testbeds and discuss the features and limitations that we found during our experiments.

As intelligent knowledge-based decision makers, Machine Learning (ML) end applications highly depend on input data quality. Systems that integrate ML-end applications use computer networks for data delivery from the data source to ML processors. Packet losses and lack of resources are the factors that may result in a quality degradation of data transmitted over computer networks, and consequently negatively impact the accuracy of ML applications. In this paper, we investigate the relation between network quality of service (QoS) degradation and ML-end decision making performance based on the data transmitted over various network conditions. For realistic testing, we implement an empirical study, in which we leverage the POWDER platform that provides real-world wireless 5G network testbed. In our experiments, we transfer media-files between the nodes of this real network under various QoS conditions. We investigate the effect of degraded data quality due to those network disruptions on ML-based systems used in different domains. Specifically, we investigate the accuracy of different image classification and speech transcription models in scenarios with higher packet loss in the communication channel and resource shortage on the receiving end. Our results show that network conditions stability is critical to avoid ML misclassifications, and that data quality degradation affects speech transcription model performance to a higher degree compared to considered image classifiers.

Internet of Things (IoT) evolution calls for stringent communication demands, including low delay and reliability. At the same time, wireless mesh technology is used to extend the communication range of IoT deployments, in a multi-hop manner. However, Wireless Mesh Networks (WMNs) are facing unstable topologies and link failures leading to unsatisfied IoT requirements. Named-Data Networking (NDN) can enhance WMNs to meet these IoT requirements, due to the content naming scheme and in-network caching, but necessitates adaptability to the challenging conditions of WMNs. In this work, we argue that Software-Defined Networking (SDN) is an ideal solution to fill this gap and introduce an integrated SDN-NDN deployment over WMNs involving: i) global view of the network in real-time; ii) centralized decision making; and iii) dynamic NDN adaptation to network changes. The proposed system is deployed and evaluated over the w-iLab.1 FED4FIRE test-bed. The proof-of-concept results validate that the centralized control of SDN effectively supports the NDN operation in unstable topologies with frequent dynamic changes, such as the WMNs.

Network experiments using real hardware are typically expensive and time-consuming. Multiple solutions exist to reduce costs, in particular network emulation, or simulation of performance metrics. However, each solution impacts the quality of experimental results, in particular concerning realism and precision. We propose HVNet, a novel approach to create virtualized topologies on a single host utilizing real networking hardware. Relying on real hardware, our approach offers realistic network behavior and high-precision measurements. HVNet enables measurements on flexible network topologies avoiding the drawbacks of the alternative solutions. We observed repeatable results with a small error margin and a low impact of the measurement setup on experimental results. Additionally, we compare latency and jitter distributions of HVNet and Mininet setups, observing an improvement factor of up to three orders of magnitude.

Monitoring Fog infrastructures in a lightweight and fault-resilient manner is an open and challenging research problem. In this article, we illustrate the experimental assessment of a distributed, self-organising and fault-tolerant monitoring tool especially targeting Fog infrastructures. The assessment involved up to 40 nodes across two testbeds within the Fed4Fire+ federation. Results show the capability of the tool to handle different types of failures in the monitored infrastructure, and quantify its measurement accuracy and limited footprint.

Current trends in Unmanned Aircraft Systems (UAS) aim at embedding intelligent functionality into these devices to enhance their utilization beyond their traditional video monitoring and recording operations, using them as complex mobile computing platforms or to support flexible communication network infrastructures. As a consequence, while UAS necessarily increase their complexity, the whole development cycle of services/applications of UAS, together with the management, operation, testing, validation, and maintenance, becomes even more challenging since the computing hardware has to be onboarded and ready to fly. This article presents a Medium Altitude Long Endurance (MALE) surveillance UAS developed by the Spanish Ministry of Defense (the MILANO, from the National Institute of Aerospace Technology), used as a reference for the developments in this paper. The article also presents the virtualization platform that is being used in the system to facilitate the deployment of all the communication service components in the real MILANO platform (e.g., routing, switching, channel selection functionalities), although it also supports other types of applications (e.g., telemetry preprocessing, including sensors and video, route planning). In addition, this platform is complemented by an emulation infrastructure (to reproduce different mobility patterns and radio communication link status) used to boost the whole testing and validation cycle before onboarding the hardware and scheduling expensive flight campaigns, and facilitating the overall UAS software maintenance during the operations phase. Finally, the article describes the experiments performed with the MILANO equipment, validating the application and functionality of the overall platform.

For efficient use of Massive MIMO systems, fast and accurate channel estimation is very important. But the Large-scale antenna array presence requires high pilot overhead for high accuracy of estimation. Also, when used with software-based processing systems like CPUs and GPUs, high processing latency becomes a major issue. To reduce Pilot overhead, a Pilot transmission scheme in combination with PN Sequence correlation based channel estimation scheme is implemented. Then, to deal with the issue of high processing latency, Tensor Cores in Nvidia GPUs are used for computing the channel estimation. Experiments are performed by using Nvidia V100 GPU in the ORBIT Testbed to show the performance of the Pilot transmission scheme. By varying factors like PN sequence length, Channel Impulse Response length, number of multiplexed transmitters, and scale of MIMO, the accuracy and processing latency of Tensor Core implementation of the Channel Estimation is evaluated.

This talk does not have an abstract.

This talk does not have an abstract.

This talk does not have an abstract.