Demo Sessions

DM 12: MENTORED: The Brazilian Cybersecurity Testbed
DM 14: YouTube goes 5G: Benchmarking YouTube in 4G vs 5G Through Open Datasets
DM 18: UAVs Allocation and Visualization in VANETs via DRL
DM 7: Improving the DASH QoS by dropping packets in programmable data planes

DM 19: Generating Mobility-aware Traces for IoT Applications
DM 20: ERENO-UI: A Tool for Generating IEC-61850 Intrusion Dataset

 

Christoph Müller, Fraunhofer FOKUS, Germany

 

Video streaming content differs in terms of complexity and requires title-specific encoding settings in order to achieve a certain visual quality. As such, a per-title encoding solution was developed in 2015 in order to identify the optimal encoding settings for a single asset, based on a prior complexity analysis of the title. The most common approach in determining a title's complexity is done by performing multiple test-encodes and calculating a quality metric like VMAF or PSNR. Based on the resulting bitrate/quality pairs, content specific bitrate/resolution ladders are derived and applied. However, a major downside of this approach is that deriving test encodes is computationally heavy, thus time consuming and not particularly applicable to live streaming scenarios. In this demo, we introduce a content aware encoding workflow that automates the standard per-title encoding method with the support of machine learning models, while avoiding the computationally heavy test encodes. The models are trained on complex datasets composed of 30+ video attributes and generate an optimal encoding ladder as an output (bitrate/resolution pairs). In comparison to the conventional per-title encoding method, we observed significant savings in terms of storage and delivery costs, while maintaining the same visual quality.

Adam Flizikowski, IS-Wireless, Poland

Evgeniy Alkhovik, IS-Wireless, Poland
Md Arifur Rahman, IS-Wireless, Poland

 

Abstract: 
This demo addresses the 5G open-RAN mobile network deployment using Kubernetes cluster on top of the edge micro data center (EMDC). It demonstrates the successful deployment of virtualized 5G network components as Kubernetes containers in just a few minutes. The cluster can run on general purpose processors architecture - i.e. edge, cloud servers or PC - however the target footprint for deployment is the AI/ML driven edge server of the EMDC. In this demo regular PCs will represent the edge-micro data center (EMDC) innovative design, which will soon be demonstrated in the BRAINE ECSEL JU research project. As regards the 5G O-RAN disaggregated RAN/Core there are no accelerators required to deploy such network, and the system is capable of monitoring various metrics in order to enable building model for prediction of future traffic demands (e.g. CPU, memory, resource block utilization, number of users etc). The 5G network used will be disaggregated based on 3GPP functional splits RU/DU/CU/core and these components are distributed over multiple computing nodes. All in all the ultimate goal behind this demo is to indicate current capabilities of the workload prediction for the RU/DU/CU components, as enabler towards the scaling of vRAN. We are utilizing the LSTM, nBeats and ARIMA models to train models for predicting the CPU consumption of disaggregated 5G. It is based on the 2 hour scenario for UE activity in the local laboratory and the time series database. The telemetry system of the EMDC monitors the consumption of resources (e.g., CPU, RAM, and storage) with regards to computing and RAN resources. After the initial connection of a new UE, a telemetry system will observe resources consumed by UE and store it in a real-time database. Receiving statistics from vRAN and EMDC, the LSTM model makes a prediction of future workload and sends the output to the placement agent. Based on the predictions, the placement agent will in future trigger vRAN component scaling to address computing resource shortage. During the demo session, we will show the real 5G data call, utilizing the Kubernetes containers and deployment prepared for the EMDC federation as well as the model training in AWS.

Zijian Zhang, Tsinghua University, China

Yuhao Chen, Tsinghua University, China

Zhenchen Peng, Tsinghua University, China

Linglong Dai, Tsinghua University, China

 

Reconfigurable intelligent surfaces (RISs) have emerged as a candidate technology for future 6G networks. However, due to the "multiplicative fading" effect, the existing passive RISs only achieve a negligible capacity gain in environments with strong direct links. To tackle this challenge, in this demo, we show a new RIS architecture called active RISs. Unlike the existing passive RISs that reflect signals without amplification, active RISs can amplify the reflected signals via amplifiers integrated into their elements. To characterize the signal amplification and incorporate the noise introduced by the active components, we verify the signal model of active RISs through the experimental measurements on a fabricated active RIS element. Based on the verified signal model, we formulate the sum-rate maximization problem for an active RIS aided multi-user multiple-input single-output (MU-MISO) system and a joint transmit beamforming and reflect precoding algorithm is proposed to solve this problem. Finally, we develop a 64-element active RIS aided wireless communication prototype, and the significant gain of active RISs are validated by field test.

Francesco D'Alterio, Sapienza University of Rome & Fondazione Ugo Bordoni (FUB), Italy

Samuela Persia, Fondazione Ugo Bordoni, Italy

Sebastiano Mario Trigila, Fondazione Ugo Bordoni, Italy

Mirko Ermini, RFI, Italy

Giuseppe Cadavero, RFI, Italy

Lorenzo Santilli, TIM - Telecom Italia SPA, Italy

Giuliano Marcelli, Telespazio, Italy

Irene Facchin, Fondazione Bruno Kessler, Italy

Gianfranco Fattore, Marini Impianti Industriali, Italy

Vincenzo De Palo, MerMec, Italy

 

The Smart Predictive Maintenance Application for the railway infrastructure of a whole country needs to overcome all typical coverage issues experienced by terrestrial networks to reach high level of reliability and affordability. To this end, we propose a communication solution based on the use of both Satellite-based and 5G terrestrial access to delivery data traffic coming from railway diagnostic sensors to central data processing units, guaranteeing a significant improvement of the network coverage and outage resilience. Traffic steering is performed by Edge Nodes, capable of selecting the best suitable or maybe the only available network access resources, using information provided by telco Network Service Orchestrator, designed for 5G networks, properly embedded with a Network Adaptation Function to manage satellite resources as well. This demo presents an On-Train PoC of the architecture which puts together telco resources made available by a 5G operator and a Satellite operator, under the orchestration and virtualization framework of ETSI MANO. The platform can transmit diagnostic data to central processing unit while train is moving leveraging both 5G and SAT RAN. Edge Node selects the most suitable network anytime and establishes a GRE tunnel among parties. Some considerations and expected impact of such platform on future diagnostic railway maintenance conclude the presentation.

Junye Li, University of New South Wales (UNSW) Sydney, Australia

Aryan Sharma, University of New South Wales, Australia

Deepak Mishra, University of New South Wales (UNSW) Sydney, Australia

Joseph Davis, University of Sydney, Australia

Aruna Seneviratne, University of New South Wales, Australia

 

With the recent advancements in the field of WiFi sensing and the upcoming IEEE802.11bf SENS standard, WiFi sensing is expected to boom in the near future. Leveraging the widespread availability of WiFi infrastructure, WiFi sensing measures the multipath propagation characteristics in an environment to infer physical changes such as its occupancy. Accurate occupancy counting has been demonstrated successfully in various studies with offline machine learning on expensive computers. However, one challenge in its commercial deployment is the issue that WiFi sensing is highly environment-specific; hence the underlying training model does not remain robust in a changing environment. To this end, we demonstrate a dynamic learning framework that enables WiFi sensing platforms to adapt to environmental changes by updating the training model using camera-based labelling. Without losing generality, we showcase a real-time, indoor human counting application using standalone intelligent WiFi sensing technology on low-cost, off-the-shelf Raspberry Pi 4B devices with the integrated camera module. The demonstrated dynamic learning approach is able to achieve comparable occupancy classification performance for changing environments as an offline-trained model with a large dataset for fixed environments while maintaining low complexity and operating autonomously.

Janne Ali-Tolppa, Nokia, Finland

Nikolaj Marchenko, Robert Bosch GmbH, Germany

Patrick Agostini, TU Berlin, Germany

Rastin Pries,Nokia, Germany

Michael Baumgart, Infosim, Germany

 

In this demo, we demonstrate how machine learning based network QoS prediction can be utilized to predict the maximum achievable throughput for Automated Guided Vehicle (AGV) applications in private industrial campus networks. Furthermore, we show how the network QoS prediction can be used to optimize the edge cloud offloading orchestration decisions, i.e., the placement decisions, for a Vision-Based Positioning (VBP) algorithm. The demonstration is based on data that has been collected from a 5G test network set up in a real operational factory hall as part of the KICK project. We will also demonstrate the latest developed algorithms and the visualization of their evaluation.

Leandro Almeida, Instituto Federal de Educação, Ciência e Tecnologia da Paraíba, Brazil

Guilherme Matos, Federal University of São Carlos, Brazil

Rafael Pasquini, Federal University of Uberlândia - UFU, Brazil

Chrysa Papagianni, University of Amsterdam, The Netherlands

Fábio Luciano Verdi, Federal University of São Carlos, Brazil

 

Video services account for the largest share of all Internet traffic, demanding a network capable of supporting the requirements of delay-sensitive traffic. Fluctuations in network load can cause high delays in the queues of network routers, which tend to degrade the Quality of Service (QoS) for adaptive video streaming, such as Dynamic Adaptive Streaming over HTTP (DASH). This work is positioned in the scope of active management queues (AQM) to improve the QoS of a DASH service by means of dropping packets. One traditional AQM that adopts a packet drop policy is Random Early Detection (RED), developed to drain the flow in times of congestion and thus reduce queueing delay. We revisited and implemented a P4-based implementation of RED, named iRED (ingress RED), an algorithm capable of dropping packets at the ingress pipeline, an innovation compared to other AQM strategies based on dropping at the egress. We compare iRED versus Tail Drop (TD) approach in an emulated programmable Content Delivery Network (CDN) employing DASH. Experiments indicate that the iRED improve the QoS in terms of cached video available in the client's buffer and in Frames Per Second (FPS) played.

Merim Dzaferagic, Trinity College Dublin, Ireland

Darijo Raca, University of Sarajevo, Bosnia and Herzegovina

Bruno Missi Xavier, Federal Institute of Espirito Santo - Campus Cachoeiro de Itapemirim, Brazil

Daragh King, Trinity College Dublin, Ireland

Marco Ruffini, CONNECT, Trinity College Dublin, Ireland

 

We are optimizing the operation of the RAN by collecting measurements, predicting the future channel quality and dynamically adjusting the scheduler policies to minimize the power consumption, while maximizing the transmission reliability and QoE for video streaming. The introduction of OpenRAN allows us to bring intelligence to the very edge of the RAN and to tailor the configuration to the traffic requirements. We modified the implementation of the srsRAN virtual BS. Our modifications expose the RAN measurements to our RIC agent and at the same time allow us to change the BS configuration during runtime. We also implemented a near-RT RIC and a RIC agent. The near-RT RIC consists of a databus and a set of xApps. The databus is used for information exchange between the xApps and the RIC agent. The RIC agent runs on the BS side and allows us to expose the RAN measurements to the RIC. The measurements are sent through the databus to the subscribed xApps. The xApps use the databus to send reconfiguration commands to the RIC agent which uses the newly implemented API on the virtual BSs to dynamically reconfigure them. We demonstrate the operation of the near-RT RIC by implementing an xApp that optimizes the scheduler policy for video streaming applications, while minimizing the power consumption.

Muddassar Farooq, FAST-NU, Pakistan

Kenneth Stanwood, Wilan Research Inc, USA

Zain Noman, Pi Invent Enterprises, Pakistan

Arslan Mumtaz, Pi Invent Enterprises, Pakistan

 

Modern telecommunication devices rely on strong encryption algorithms for safe and secure communication. The cryptographic strength of an encryption algorithm can be improved by using multiple heterogeneous modules. However, designers and users are unable to evaluate the randomness added to the cipher due to individual module. Test suites do not provide any insight on the strength of the modules within an encryption algorithm. There does not exist a tool that can be used to compare the cryptographic strength of the various modules within an encryption algorithms such as S-Box, Diffusion layer, 1 Round e.t.c. CryptoView measures strength of an encryption algorithm or stream generated from a True Random Number Generator (TRNG).

Haijun Zhang, University of Science and Technology Beijing, China

Wanqing Guan, University of Science and Technology Beijing, China

Dong Wang, University of Science and Technology Beijing, China

Chunlei Sun, University of Science and Technology Beijing, China

Qize Song, University of Science and Technology Beijing, China

A Nallanathan, University of Science and Technology Beijing, China

 

 Real-time VR/AR services can efficiently support interactive 3D-views and immersive experience, and will boost the emergence of more intelligent applications in B5G and 6G eras. Our demo mainly focuses on the intelligent and efficient VR/AR in B5G and 6G networks. We virtualize the physical GPU into logical components to facilitate the flexible scheduling of computing power on demand. Then, we design a Real Time Controller (RTC), which provides customized services by invoking AI algorithms in computing power components, thus realizing intelligent and efficient VR/AR. Our demo system runs in a self-established B5G/6G private network, which can realize intelligent and real-time target detection in live VR videos, and intelligent resource management in RAN. Significantly, we have successfully achieved a detection delay lower than 10 ms, a frame rate higher than 15 fps and increased the resource utilization by 30%. Our project will promote the real applications of VR/AR in future B5G and 6G networks.

Renan C. A. Alves, Universidade de São Paulo, Brazil

Marcos A. Simplicio Jr., Universidade de São Paulo, Brazil

Bruno C Albertini, Universidade de São Paulo, Brazil

 

 Hardware level attacks are more difficult to detect in comparison to software level attacks, since traditional security tools, such as anti-virus, do not analyze low level communication and code. The Secure Protocol and Data Model (SPDM) standardizes hardware level security primitives, such as authentication, firmware measurement, and certificate management. In this demonstration, we show how SPDM can be used to secure the communication between devices and the operating system on a virtualized environment. More specifically, we adapted a VirtIO hard drive and a E1000 network card to use the standard and measured what is the expected impact on performance, using a variety of benchmark tools.

Bruno H. Meyer, Federal University of Parana, Brazil

Davi Daniel Gemmer, UTFPR, Brazil

Marcos Schwarz, RNP, Brazil

Emerson R de Mello, Federal Institute of Santa Catarina, Brazi

Michelle Silva Wangham, University of Vale do Itajaí, Brazil

 

With the increasing IoT applications and their importance to the global economy, communication infrastructures and services were an essential tool and fundamental element to public well-being and economic stability. The large scale of the Internet of Things requires complex testbeds capable of supporting experimental scenarios with sufficient scale to efficiently evaluate cybersecurity solutions against botnet-based DDoS attacks. The purpose of this demo is to present how the MENTORED Testbed can satisfy different requirements of Cybersecurity testbeds, such as real-time monitoring, user-centric perspective, scalability, and fidelity. The MENTORED Testbed architecture focuses on DDoS attacks and IoT devices, with access control policies. The Testbed can be used to create complex and custom network virtualizations using the Software-Defined Infrastructure of the National Education and Research Network (IDS-RNP) based on Kubernetes and a technology named Knetlab for network virtualization. This demo shows a rest API, enabling the definition, creation, and execution of cybersecurity experiments using only Web Browsers. We analyzed the proposed testbed behavior through a case study that reproduces the network traffic in a DDoS attack scenario that uses different Brazil regions.

Rossano P. Pinto, School of Technology of Americana - Centro Paula Souza, Brazil

Kayol Mayer, University of Campinas - UNICAMP, Brazil

Christian Esteve Rothenberg, University of Campinas - UNICAMP, Brazil

Darli Mello, University of Campinas - UNICAMP, Brazil

Dalton Arantes, University of Campinas - UNICAMP, Brazil

 

Network Slices can be used to offer premium clients several services, including a soft protection path (SPP). SPP carries traffic from both premium and non-premium clients when the working path becomes unavailable. Premium clients' SLA must be guaranteed. In SDN networks, the controller must detect the path unavailability and redirect premium traffic to the SPP reconfiguring it to guarantee the premium client's SLA. The needed redirection time is usually tied to an upper bound. A local fast response agent deals with link unavailability faster than the controller. At the same time, the local agent coordinates it's actions with the controller. This can be seen as a hierarchical control plane.

Raza Ul Mustafa, University of Campinas - UNICAMP, Brazil

Chadi Barakat, Inria, Université Côte d'Azur, France

Christian Esteve Rothenberg, University of Campinas - UNICAMP, Brazil

 

As 5G technology evolves, its performance is expected to improve over time. Therefore, the QoE of the YouTube video streaming from Mobile Network Operators (MNOs) perspective is ideal and challenging compared to 4G/LTE networks. Evaluating mobile carriers' end-to-end network performance in the wild is known to be difficult and complicated. Critical issues for MNO include how to manage increased video traffic demands and provide a satisfactory Quality of Experience (QoE) experience to their end-users. To ensure better QoE, understanding and monitoring the Key Performance Indicators (KPIs) that impact users' perceived QoE has become a trending topic. Therefore, we carry out a massive 4G and 5G dataset collection campaign using a commercial 4G and 5G network, where we consider YouTube as baseline for video streaming to collect Channel Metrics and YouTube QoE logs with 1-second granularity.

Raza Ul Mustafa, University of Campinas - UNICAMP, Brazil

Christian Esteve Rothenberg, University of Campinas - UNICAMP, Brazil

 

TLS encryption establishes a more secure and private connection, where classic Deep Packet Inspection (DPI) techniques no longer provide valuable information. Due to limited information available to inspect video flows, it is incredibly difficult to find novel lightweight QoS patterns affecting the video QoE that requires less computations and processing. The exponentially increasing video traffic demands pose many challenges for the MNOs to consistently provide maximum QoS resources for higher QoE. Traditional approaches consider YouTube as a baseline to derive QoS features from the encrypted streams along with chunk level statistics, where YouTube delivery mechanisms vary over time.

Francisco Germano Vogt, University of Campinas - UNICAMP, Brazil

Fabricio E Rodriguez Cesen, University of Campinas - UNICAMP, Brazil

Christian Esteve Rothenberg, University of Campinas - UNICAMP, Brazil

Gergely Pongrácz, Ericsson Research, Hungary

 

Francisco Germano Vogt, Fabricio E Rodriguez Cesen and Christian Esteve Rothenberg (University of Campinas - UNI CAMP, Brazil); Gergely Pongrácz (Ericsson Research, Hungary)

 

Network monitoring is a fundamental task to provide good network management and performance. Since the SDN emergence, the In-band Network Telemetry (INT) has been demonstrated as an efficient network monitoring framework. Using INT, we can collect network information hop-by-hop directly from the data plane by including this information in the network production traffic. However, this information collection is limited by available packet size and processing overhead, making it critical to choose what data to collect and when to collect it. So, in this work, we propose the Inter-Packet Gap (IPG) per-hop monitoring using INT. We argue that by monitoring the IPG hop-by-hop, it is possible to correlate the data and identify network problems like network congestion, delay, and microbursts and their contributing flows. Our preliminary results show that IPGNET can efficiently detect the microbursts on multiple queues and report all the contributing flows.

Fabricio E Rodriguez Cesen, University of Campinas - UNICAMP, Brazil

Géza Szabó, Ericsson Research, Hungary

Christian Esteve Rothenberg, University of Campinas - UNICAMP, Brazil

Gergely Pongrácz, Ericsson Research, Hungary

 

With 5G networks, new opportunities and challenges appear just around the corner. Applications in different trends (e.g., industry 4.0, agriculture, IoT, UAV) are gaining new envisions. The network's capabilities need to grow together with the high amounts of transmitted data and various network services. Supporting Ultra-low latency (ULL) applications with extremely low loss and delay variation is required. Despite the significant characteristics of the 5G networks, horizontal (e.g., distance, nodes processing) and vertical (e.g., NIC, OS, Hypervisor, Application) delays affect the performance of end-to-end communication. Unmanned aerial vehicles (UAVs), especially drones, are becoming an important application powered by the new network characteristics. We propose an in-network UAV collision avoidance algorithm in edge Programmable Data Planes. We explore the benefits of an edge P4-based device (e.g., High Performance, Reconfigurability, Protocol Independent) in a ULL use case (i.e., UAVs). The collision avoidance algorithm implemented in P4 can detect and fast-react to avoid possible impacts effectively.

William Quintas de Melo, University of Campinas - UNICAMP, Brazil

Christian Esteve Rothenberg, University of Campinas - UNICAMP, Brazil

 

Vehicular Ad hoc Networks require its resources be fairly distributed across its connected resources. However, the terrestrial telecommunications infrastructure doesn't ever ensure that it happens. The usage of Unmanned Aerial Vehicles (UAVs) as support resources is a promising technology to the future of wireless communications, specially with the 5G being launched. The UAVs positioning that maximizes this items contribution to the network is being studied in many researches. This work presents a simulation platform to help in these studies, integrating wireless networks emulator Mininet-WiFi and Reinforcement Learning library OpenAI Gym.

Kevin Jiokeng, Inria Lille, France

Nina Santi, INRIA, France

Nathalie Mitton, Inria Lille - Nord Europe, France

 

This demonstration introduces a set of tools that enables the reproducible generation of traces for IoT applications taking into account mobility and fine-grained instantaneous energy consumption information. Our setup leverages the FIT IoT-Lab open testbed and its hundreds of nodes to enable researchers to remotely build datasets for their custom IoT/Edge scenarios. We demonstrate this through a sample scenario composed of 100 IoT nodes belonging to 3 mobility-aware applications and make the resulting datasets available to the community. The generation of these kinds of traces is important for designing and evaluating accurate mobility aware offloading algorithms for IoT and Edge Computing and, more specifically, for training Artificial Intelligence models.

Silvio E Quincozes, Universidade Federal de Uberlândia, Brazil

Vagner Quincozes, Universidade Federal do Pampa, Brazil

Célio Vinicius Neves de Albuquerque, Universidade Federal Fluminense, Brazil

Diego Passos, Universidade Federal de Uberlândia, Brazil

Daniel Mosse, University of Pittsburgh, USA

 

Digital substation networks based on IEC-61850 standards are vulnerable to several attacks. Therefore, employing Intrusion Detection Systems (IDSs) is fundamental. However, IDSs for digital substation are still at an early stage. In this demo, we propose ERENO-UI to enable the easy generation of datasets for Machine Learning-based IDSs.