Who?
Names:
COATI Chuan Xu, Frédéric Giroire
EPIONE Marco Lorenzi
NOKIA BELL LABS Marie-Line Alberi-Morel
Mail: chuan.xu@inria.fr, Frederic.giroire@inria.fr,marco.lorenzi@inria.fr
Where?
Place of the project: Inria Sophia Antipolis
Address: 2004 route des Lucioles
Team: COATI and EPIONE Teams
Web pages:
https://team.inria.fr/epione/en/
Pre-requisites if any: The candidate should have a solid mathematical background and/or good programming skills and previous experience with PyTorch or TensorFlow.
Description:
Context
Future sixth-generation (6G) networks will be highly heterogeneous, with the massive development of mobile edge computing inside networks. Furthermore, 6G is expected to support dynamic network environments and provide diversified intelligent services with stringent Quality of Service (QoS) re- quirements. Various new intelligent applications and services will emerge (including augmented reality (AR), wireless machine interaction, smart city, etc) and will enable tactile communications and Internet of everything (IoE). This will challenge wireless networks in the dimensions of delay, energy consumption, interaction, reliability, and degree of intelligence and knowledge, but also in the dimension of information and data sharing. In turn, 6G networks will be expected about leveraging data at the next step of the new communication system generation. First of all, they will generate large amounts of data much more data than 5G networks: multiple sources as Core, Radio Access Network, OAM, User Equipments (UEs) but also as private and/or personal devices/machines massively con- nected, data-generator applications as sensing, localization, context-awareness services etc. Besides, unlike today’s networks where traffic is almost entirely centralized, most 6G traffic will remain localized and highly distributed. The communication system will not only provide the bits reliably, but more importantly will provide the intelligent data processing through connectivity and resources computing in the devices, the edge, and the cloud in the network. For this, with Artificial Intelligence (AI) and Machine Learning (ML), machines will bring to networks the necessary intelligence very close to the place of action and decision-making and will also make data sharing possible.
Reliable and efficient transmission, data privacy and security are great challenges in data sharing. Specially for 5G advanced and 6G networks data is distributed with the wide deployment of various connected Internet of Thing (IoT) devices, and are generated from many distributed network nodes, e.g., end users, small Base Stations or Distributed Units and the network edge. Also, how to collect/share efficiently data from multiple sources (e.g., sensors or device) up to AI/ML-based Network applications/services of Orchestration and Automation Layer (network management system) in Edge? The models shall be trained, updated regularly and operate in real-time.
Recently, generative models have been demonstrated playing a key role in data sharing while pre- serving privacy and security. They are able to generate synthetic data which distribution is similar to the original data one. So, instead of sending original data, many applications (medical or financial) use them to transfer data. Generative models are shown be useful in many scenarios such as health and financial applications [VSV+22]. However, the highly distributed architecture in 5G advanced/6G motivates the need for distributed, multi-agent learning for building generative models located at given anchor points of data collection (Edge server or Central Units) inside the RAN/Edge.
Challenges and objectives
We aim to design a communication-efficient and privacy-preserving on demand framework such that the local agents inside RAN/Edge cooperatively generate a synthetic dataset which represents well the global data distribution for model utility. To this end, one can train a generative adversarial network in a federated way [AMR+20], where the agents and the server alternatively minimize the loss function of the discriminator and the generator. However, deep generative models have a tendency to memorize the training examples which may leak private information [HMDC19, CYZF20]. While, applying the traditional privacy-preserving defense such as differential privacy mechanism [Dwo06] will degrade the generative model’s utility and thus influences the synthetic data quality. Moreover, the training requires 500-10000 communication rounds in practice for convergence (see [KMA+21, Table 2]) which is expensive for communication cost. Recently, there is another work [ZCL+22] where the server makes uses of all the local trained models to train a generator, which minimizes the communication cost to only one round. However, transferring these local models are extremely dangerous as they can be used to infer the private information on the dataset of devices [FJR15, YGFJ18]. Alternatively, instead of transferring the models as the previous work proposed, the devices can transfer directly the distilled synthetic data which are computed locally [ZPM+20]. However, the quality of the assembled synthetic dataset degrades especially when some agents have just few training samples.
The goal of the project will be to compare the above-mentioned existing methods for synthetic dataset generation, in terms of their trade-offs on model accuracy, data similarity, communication cost, model compression and privacy.
The project is part of a larger collaborative project with Nokia Bell Labs Core Reasearch.
A PhD grant is funded as part of the global project.
The TER can thus be followed by a PhD for a motivated student.
References
[AMR+20] Sean Augenstein, H Brendan McMahan, Daniel Ramage, Swaroop Ramaswamy, Peter Kairouz, Mingqing Chen, Rajiv Mathews, et al. Generative models for effective ml on private, decentralized datasets. ICLR, 2020.
[CYZF20] Dingfan Chen, Ning Yu, Yang Zhang, and Mario Fritz. Gan-leaks: A taxonomy of member- ship inference attacks against generative models. In Proceedings of the 2020 ACM SIGSAC conference on computer and communications security, pages 343–362, 2020.
[Dwo06] Cynthia Dwork. Differential privacy. In Automata, Languages and Programming: 33rd International Colloquium, ICALP 2006, Venice, Italy, July 10-14, 2006, Proceedings, Part II 33, pages 1–12. Springer, 2006.
[FJR15] Matt Fredrikson, Somesh Jha, and Thomas Ristenpart. Model inversion attacks that exploit confidence information and basic countermeasures. In Proceedings of the 22nd ACM SIGSAC Conference on Computer and Communications Security, pages 1322–1333, 2015.
[HMDC19] Jamie Hayes, Luca Melis, George Danezis, and Emiliano De Cristofaro. LOGAN: mem- bership inference attacks against generative models. Proc. Priv. Enhancing Technol., 2019(1):133–152, 2019.
[KMA+21] Peter Kairouz, H Brendan McMahan, Brendan Avent, Aur ́elien Bellet, Mehdi Bennis, Ar- jun Nitin Bhagoji, Kallista Bonawitz, Zachary Charles, Graham Cormode, Rachel Cum- mings, et al. Advances and open problems in federated learning. Foundations and Trends® in Machine Learning, 14(1–2):1–210, 2021.
[VSV+22] Rohit Venugopal, Noman Shafqat, Ishwar Venugopal, Benjamin Mark John Tillbury, Harry Demetrios Stafford, and Aikaterini Bourazeri. Privacy preserving generative adver- sarial networks to model electronic health records. Neural Networks, 153:339–348, 2022.
[YGFJ18] Samuel Yeom, Irene Giacomelli, Matt Fredrikson, and Somesh Jha. Privacy risk in machine learning: Analyzing the connection to overfitting. In 2018 IEEE 31st Computer Security Foundations Symposium (CSF), pages 268–282. IEEE, 2018.
[ZCL+ 22] Jie Zhang, Chen Chen, Bo Li, Lingjuan Lyu, Shuang Wu, Shouhong Ding, Chunhua Shen, and Chao Wu. Dense: Data-free one-shot federated learning. In Advances in Neural Information Processing Systems, 2022.
[ZPM+ 20] Yanlin Zhou, George Pu, Xiyao Ma, Xiaolin Li, and Dapeng Oliver Wu. Distilled one-shot federated learning. ArXiv, abs/2009.07999, 2020.
Who?
Name: Sid TOUATI
Mail: Sid.Touati@inria.fr
Where?
Place of the project: INRIA Sophia Antipolis
Web page: http://www-sop.inria.fr/members/Sid.Touati/
What?
Pre-requisites if any: Compilation, informatique fondamentale
Detailed description: indicate the context of the work, what is expected from the intern, what will be the outcome (software, publication, ...).
Expected: software
References: set of bibliographical references (article, books, whitepapers, etc) to be read by the student before starting to work on this subject)
Voir fichier pdf joint.
Poursuite en thèse possible.
Who?
Name: Chadi Barakat and Thierry Turletti
Mail: Chadi.Barakat@inria.fr and Thierry.Turletti@inria.fr
Telephone: 04 92 38 77 77
Web page: https://team.inria.fr/diana/chadi and https://team.inria.fr/diana/team-members/thierry-turletti/
Where?
Place of the project: Diana Project-Team, Inria centre at Université Côte d'Azur
Address: 2004, route des lucioles, 06902 Sophia Antipolis, France
Team: Diana team
Web page: https://team.inria.fr/diana/
What?
Pre-requisites if any: Strong knowledge in network protocols, mobile networks, network measurement, data analytics. Strong programming skills: python, scripting, java/C++, etc.
Detailed description:
Mobile networks (5G and beyond) are witnessing a revolution nowadays with the increase in the bitrate, the densification of the wireless cells, and the advent of virtualization and softwarization allowing to deploy network functions and services in data centres, placed at the edge of the network. The efficient management of the network resources (both wireless and wired), and the optimal placement and orchestration of communications between virtual functions and with the users, requires the deployment of a monitoring plane allowing to discover and profile the available resources at the edge in real time and, in parallel, provide stakeholders (operators, providers and end users) with a sufficient level of information on the capacity and connectivity of these resources so that everyone can collaborate for improved network resource consumption and optimized Quality of Experience to end users. An efficient network management that takes into account the quality of experience of end users does not only depend on simple network metrics such as the delay, or physical proximity, but rather on a complex set of metrics such as the bitrate in both directions, the jitter, the packet loss rate, the context of mobility, the device properties, etc. The collection of all these metrics in an accurate and timely way represents a real challenge. Further, and given the large number of devices foreseen at the edge and their mobility and time dynamics, the measurement plane has to be of low cost, able to scale with the number of users, devices and services, and must track the whole system in an efficient manner. This is another challenge facing the development of a monitoring plane for future mobile networks. The analysis of the collected data is also a challenge as it will be rich and diverse, not necessarily directly explainable and related to network events, so new data-driven algorithms based on machine learning and artificial intelligence are needed to process the collected data, interpret it in terms of user quality of experience and network performance, and transform it into indicators on what can be the root cause of any service degradation.
The objective of our activity in the Diana team is to contribute to the development of this monitoring plane. Our work is funded by the French National Project on 5G networks and networks of the future (PEPR 5G). First, we will go over the existing solutions for mobile edge monitoring, and benchmark them to evaluate their capacity to accommodate the requirements of different applications. We will focus in particular on how information is collected, what information is collected, and to which extent existing solutions can improve the service optimization at the edge by taking into account the requirements of applications and the amount of available resources, both in the network and the computing infrastructure. For this study, we will follow an experimental approach and deploy scenarios over wireless platforms such as Mininet WiFi, R2lab wireless platform or the SophiaNode platform.
Departing from the above results, we will move to work on a new flexible monitoring solution able to cover the different parts of a mobile edge system such as the wireless channel and the edge cloud part, and to customize the level of information provided based on the requirements of the network management plane. This new solution will mainly consist on bridging the wireless channel view with the cloud and network view in one view able to shed light on the different events that can happen inside the network and on the computing nodes of the edge cloud. We will devise experimentation scenarios to prove the flexibility of our monitoring solution and its efficiency to accommodate the different monitoring requests.
This internship proposal fits within this roadmap. It will be followed by a PhD thesis if clear motivation and satisfactory results (secured funding by PEPR 5G). We will consider the RNIS tool for the monitoring of the 5G radio interface, and the prometheus/grafana tools for the network and cloud part, deployed them in R2Lab, overview and understand the details of their measurements, and establish models, following an approach based on controlled experimentation and machine learning, to bridge the gap between these measurements and the quality of experience of the user. We will consider applications like video streaming and web browsing for the purpose of this study. The internship will also seek the federation of these tools in one unique tool that forms the basis for our future monitoring plane.
References:
- Mininet WiFi, Emulation Platform for Software-Defined Wireless Networks, https://mininet-wifi.github.io/
- R2lab anechoic chamber, https://r2lab.inria.fr/
- M. Lahsini, T. Parmentelat, T. Turletti, W. Dabbous and D. Saucez, "Sophia-node: A Cloud-Native Mobile Network Testbed," 2022 IEEE Conference on Network Function Virtualization and Software Defined Networks (NFV-SDN), Phoenix, AZ, USA, 2022.
- S. Arora, P. A. Frangoudis and A. Ksentini, "Exposing radio network information in a MEC-in-NFV environment: the RNISaaS concept," 2019 IEEE Conference on Network Softwarization (NetSoft), Paris, France, 2019.
- Promotheus, an open-source systems monitoring and alerting toolkit, https://prometheus.io/
- Muhammad Jawad Khokhar, Thibaut Ehlinger, Chadi Barakat, “From Network Traffic Measurements to QoE for Internet Video“, in proceedings of IFIP Networking, Warsaw, Poland, May 2019.
Name: Nicolas Nisse
Mail: nicolas.nisse@inria.fr
Web page: Nicolas Nisse
Co-advisors : Frederic Giroire (frederic.giroire@inria.fr)
Place of the project: Inria Sophia Antipolis
Address: 2004 route des Lucioles, 06902 Sophia Antipolis
Team: COATI
Web page: Coati
Description:
The goal of the internship is to develop methods to analyse the evolution over time of a social network. We will consider as example the graph of scientific collaborations as it can be crawled freely.
The internship will have two phases:
Data collection. In the first phase, the student will use the available bibliographic research tools (SCOPUS, Web of Science, Patstat) to create data sets. One corresponding to the current situation and others corresponding to past moments. The data sets will correspond mainly to networks (annotated graphs) of scientific collaborations.
Data analysis. In the 2nd phase, the student will analyse this data. First, they will focus on simple metrics (number of publications, number of patent applications...) and compare the evolutions across time. Then, if there is time, she will start studying the evolution of the structure of the network and will look at whether they are observing an evolution of its clustering due to the emergence of new collaborations. The project will be part of a larger project on the evaluation of the impact of funding on scientific research. The project involve researchers in economics, sociology, and computer science.
Keywords: graph algorithms, big data, graph algorithms, network analysis
Useful Information/Bibliography: The project is the continuation of the internship whose report can be found here
Who?
Name: Chadi Barakat and Thierry Turletti and Walid Dabbous
Mail: Chadi.Barakat@inria.fr and Thierry.Turletti@inria.fr and Walid.Dabbous@inria.fr
Telephone: 04 92 38 77 77
Web page: https://team.inria.fr/diana/
Where?
Place of the project: Diana Project-Team, Inria centre at Université Côte d'Azur
Address: 2004, route des lucioles, 06902 Sophia Antipolis, France
Team: Diana team
Web page: https://team.inria.fr/diana/
What?
Pre-requisites if any: Strong knowledge in network protocols, mobile networks, network measurement, data analytics. Strong programming skills: python, scripting, C/C++, etc.
Detailed description:
With the advent of softwarization in networks, and in next generation cellular networks in particular, the current trend is to validate network solutions over emulated testbeds that have the advantage to be flexible and easily deployed. The emulation can be done either on one physical machine like Mininet, or on a cluster of physical machines like Maxinet and Distrinet. The main challenge with network emulation is to make sure that the emulation has well passed, and was not bottlenecked by the underlying network conditions or the compute resources. Realism (or fidelity) of an emulation is a sufficient condition for reproducibility of the experiment.
Network emulation can be disturbed by several phenomena such as a saturation of the underlying network, or a lack of computing resources on the physical machine(s). Verifying if an emulation has well passed is a challenging task as there is no a priori knowledge on what should be the output of the experiment itself. Monitoring the underlying infrastructure can bring some hints, but in many cases such monitoring is not made possible to the experimenter (case of a cloud experiment) and even when made possible, it does not involve a direct link between the infrastructure performance metrics and the experiment itself (the experiment itself can cause congestion of the infrastructure, which is deemed to be normal). In another context in the Diana team, we are working on a framework for emulation validation of wired networks using packet-level measurements at the emulated link level. We aim in this project, funded by the national PEPR 5G project on 5G networks and future networks, on extending this framework to sliced cellular networks embracing different technologies (edge, core) and actors (e.g. operators, cloud providers), and validating its performance in detecting and troubleshooting emulation anomalies. These heterogeneous cellular networks include a higher number of details as compared to wired networks given the complexity of their wireless part (e.g., shared medium, multi-path fading), and they are also subject to more perturbing phenomena such as the interference of other wireless devices. Moreover, the experimental evaluation of disaggregated sliced radio networks involves several components that can also be emulated such as a gNodeB physical layer or a large number of UEs for scalability. Our long term objective is to propose and implement a framework for emulation realism verification that allows (i) to establish reference models of what should be the behavior of a cellular experiment, (ii) to collect measurements about the emulation and build performance metrics that can be compared to their reference values, (iii) detect if the emulation has encountered any problem and identify the parts of the network that are responsible of the degradation, and (iv) propose solutions to remedy from the emulation problems. We will work on testing the proposed framework over wireless platforms, and in particular, the SophiaNode platform, based on R2lab, which allows running reproducible experiments in an anechoic environment.
This internship proposal fits within this roadmap. It will be followed by a PhD thesis if clear motivation and satisfactory results (secured funding by the PEPR 5G project). We will start by building on our previous work by reproducing our previous results in the context of a sliced 5G mobile network. We will overview the literature for models for the QoS promises of sliced wireless networks and devise new methods to implement them in an emulation and new metrics that we can collect to verify if the implementation is well verified or not. We will then run experiments to evaluate the real performance of these networks and check if our method is able to correctly detect any performance issue throughout the experiment. Along this work, we will identify what is needed for an efficient emulation and verification of slicing in 5G network in scenarios with multiple operators and multiple tenants.
References: set of bibliographical references (article, books, white papers, etc) to be read by the student before starting to work on this subject
- Houssam ElBouanani, Chadi Barakat, Walid Dabbous, Thierry Turletti, “Delay-based Fidelity Monitoring of Distributed Network Emulation“, in proceedings of IEEE COMSNETS Workshop on Testbeds for Advanced Systems Implementation and Research (TASIR), Bangalore, India, January 2023.
- Houssam ElBouanani, Chadi Barakat, Walid Dabbous, Thierry Turletti, “Passive Delay Measurement for Fidelity Monitoring of Distributed Network Emulation“, in Elsevier Computer Communications Journal, vol. 195, pp. 40-48, November 2022.
- R2lab anechoic chamber, https://r2lab.inria.fr/
- M. Lahsini, T. Parmentelat, T. Turletti, W. Dabbous and D. Saucez, "Sophia-node: A Cloud-Native Mobile Network Testbed," 2022 IEEE Conference on Network Function Virtualization and Software Defined Networks (NFV-SDN), Phoenix, AZ, USA, 2022.
Who?
Name: Arnaud Legout (DIANA) and Frédéric Giroire (COATI)
Mail: arnaud.legout@inria.fr and frederic.giroire@inria.fr
Telephone: +33 4 92 38 78 15
Web page: http://www-sop.inria.fr/members/Arnaud.Legout/
https://www-sop.inria.fr/members/Frederic.Giroire/
Where?
Place of the project: Inria Sophia Antipolis
Address: 2004 route des Lucioles
Team: DIANA and COATI
Web page: https://team.inria.fr/diana/
Pre-requisites if any: Python, graph theory, blockchain and privacy course, highly motivated
What?
Bitcoin is the first cryptocurrency blockchain in terms of valuation (+700B USD), it is also the most studied one. However, the internal structure of the graph of transactions is still poorly understood. Classical power law models that apply to social networks do not work for Bitcoin because the arrival date in the system is not simply correlated to the activity or popularity.
The goal of the internship is to explore the internal structure of the transaction graph for largest strongly connected component. You will have the unique opportunity to access pre-computed data structures and graphs that are otherwise very hard to collect and construct. You will have to propose interesting metrics to compute on the graph, implement these metrics, and compute them of the graph. Then it will be possible to propose models explaining the core characteristics of the graph, and in particular the main features explaining its evolution with time.
This subject will be refined with the intern and we have flexibility to move the cursor between practice and theory. For excellent contributions, we might envision an academic publication.
Useful Information/Bibliography:
This internship could continue with a Ph.D. for excellent students. For any additional information on the subject (and possible Ph.D. continuation), it is best to directly contact the advisors.
Names: Frederic Giroire and Joanna Moulierac (COATI), Guillaume Urvoy-Keller (SigNet)
Mail: frederic.giroire@inria.fr, joanna.moulierac@univ-cotedazur.fr, guillaume.urvoy-keller@univ-cotedazur.fr
Place of the project: Inria, Sophia Antipolis, 2004 route des Lucioles
Team: COATI
Webpage: Coati
Description:
With the current energy crisis, the skyrocketing cost of energy, and the global awareness of the consequences of carbon emissions on our planet, it is crucial to reduce the global energy consumption related to Information and Communication Technologies, and in particular the one related to networks. Indeed, while being commercialized worldwide, the 5G technology coupled with ultra-high resolution video has been blamed for its high energy consumption.
To answer this issue, industrials, and researchers have started to look beyond 5G to define the next 6G with at its core the need to evolve towards greener networks. The 6G standard imposes a transmission energy efficiency as one of its Key Performance Indicators (1 picoJ/bit). However, at the same time, 6G will introduce several technological breakthroughs, specifically the integration of traditional terrestrial mobile networks with emerging space, aerial, and underwater networks to provide anytime, anywhere, network access. Another critical paradigm of 6G is the utilization of Artificial Intelligence (AI) techniques to provide context-aware information transmissions and personal customized services, as well as to realize automatic network management. The growing ICT infrastructure, exploding data, and the AI-based services will result in surging energy consumption.
Thus, succeeding in the challenge of developing more energy efficient networks will require significant improvement in several directions: e.g. relaunching measurement campaigns, rethinking protocols, developing energy-efficient network management algorithms, adopting energy harvesting techniques, deciding if and when data should be sent, but also creating tools to allow the individuals to take informed decisions and have a sustainable Internet usage, in particular with the high definition video traffic.
In this internship, the student will try to refine existing power models as they are the foundations to discuss energy efficiency. She/he will make a special focus on video services, as video represents the majority of the Internet traffic and drives a significant share of the investment in terms of data centers and Internet links. She/he will start from the methods proposed in related works to estimate the carbon footprint of a typical streaming service. Some applications of 5G+/6G services in industries (e-health, hospitality, education, and gaming, smart factory, digital twin) are expected to be enabled by high data throughput, low latency and high reliability provided by future connectivity solutions. The next step will consider applications such as digital twin and smart factory.
The internship can thus be followed by a PhD for a motivated student.
Who?
Name: Arnaud Legout (DIANA)
Mail: arnaud.legout@inria.fr
Telephone: +33 4 92 38 78 15
Web page: http://www-sop.inria.fr/members/Arnaud.Legout/
Where?
Place of the project: Inria Sophia Antipolis
Address: 2004 route des Lucioles
Team: DIANA
Web page: https://team.inria.fr/diana/
Pre-requisites if any: Python, data science, statistical background (classical statistical tests,
bootstrap, resampling techniques), eager to work at the frontier of computer science,
statistics, and treatment evaluation.
What?
Medical studies have been revolutionized by the introduction of double-blind randomized
placebo-controlled trials in the 1950s and the adoption of rigorous statistical tests to
conclude on the efficacy of a treatment. This strong methodology has been the gold
standard for more than 50 years. However, there is today growing concerns about
limitations of this methodology. In particular, the ubiquitous usage of statistical tests is highly
criticized with a strong recommendation to simply drop them [1][2]. The argument of the
detractor of statistical tests is that they are abused to make erroneous conclusion on the
efficacy of a treatment. The main reasons of poor use of statistical tests are twofold: they
are hard to understand and the hypotheses (in particular normality of the distribution) are
impossible to validate.
However, there are today better alternatives to these parametric tests: resampling
techniques [3][4][5]. They are non-parametric, easy to understand and interpret, easy to
compute, and valid for numerous statistics. However, their usage is extremely rare in
medical studies as there is no study demonstrating their merits in this context.
The objectives of this internship are the following
1) Develop a methodology to evaluate the correctness and robustness of resampling
techniques on real datasets (possibly comparing the results with classical statistical
methods). This methodology will have to explore how sampling a real dataset and how
resampling a sample impact the results. The main objective is to go beyond perturbation
of a normal distribution by working on real datasets. We will also have to consider
numerical issues (such as rounding issues) on the outcome of the resampling.
2) Reevaluate published results having open data with resampling techniques and discuss
the differences. In particular, we will also explore whether the assumptions of the
classical statistical tools have been validated.
This internship is research oriented. The required qualities of the candidate are curiosity, eagerness to
challenge themselves on a difficult research subject, autonomy, flexibility, and pro-activity. The subject
is flexible and will be refined with the candidate. The ultimate goal for high quality results will be to
submit them to a top-notch international conference or journal. Students are highly encouraged to
contact Arnaud Legout (arnaud.legout@inria.fr) prior to applying.
Useful Information/Bibliography:
This internship could continue with a Ph.D. for excellent students. For any additional information on the subject (and possible Ph.D. continuation), it is best to directly contact the advisor.
[1] Retire statistical significance. Valentin Amrhein et al., Nature 2019
[2] Statistical tests, P values, confidence intervals, and power: a guide to misinterpretation. S.
Greenland et al., Eur J Epidemiol (2016)
[3] An Introduction to the Bootstrap. Bradley Efron and Robert J. Tibshirani, 1993
[4] Computer age statistical inference. Bradely Efron and Trevor Hastie, 2021
[5] The introductory Statistics Course: A ptolemaic Curriculum? George W. Cobb, Technology
Innovations in Statics Education, 2007
Names:
COATI Joanna Moulierac, Frederic Giroire
DIANA Chadi Barakat, Thierry Turletti
Mail: joanna.moulierac@inria.fr,frederic.giroire@inria.fr,chadi.barakat@inria.fr,thierry.turletti@inria.fr
Place of the project: Inria Sophia Antipolis
Address: 2004 route des Lucioles
Team: COATI and DIANA Teams
Web pages: Coati and Diana
Description:
The upcoming 5G and beyond networks will be enabler for a plethora of new applications that are known to be computing intensive, bandwidth greedy and delay sensitive, such as video streaming, virtual reality, gaming, autonomous driving, and video surveillance. The resource and energy consumption of these applications will depend on their demand and design but also on the way they are managed within the network (e.g. where to put the computing and network functions, the management of communication patterns, the coding/transcoding for video and audio). This management will determine the amount of resources utilised and energy consumed, but will also determine the Quality of Experience perceived by the users of an application. The challenge here is to manage the network in an efficient and intelligent way taking into account the requirements of the application while utilising the available resources in the most efficient way. This will first pass by modelling the computing and network requirements of new applications in the context of 5G networks, and then will rely on establishing the link between their requirements and needs, and the perceived Quality of Experience by the end users. We propose to establish this link by experimenting with an application in various network setups and various loads, then building models relating resource utilisation to network traffic following a machine learning and data-driven approach. The next phase will be to showcase how the built models can help better tune the network to efficiently utilise its resources, and consequently the consumed energy, without sacrificing the end users Quality of Experience.
In this internship, we will start exploring this topic by focusing on the particular case of gaming and/or virtual reality, browse the literature for relevant work on modeling their quality of service and quality of experience, and design a first experimental setup to run the application in a controlled environment. Our aim is to establish a list of network requirements for this application, that we can later use for function placement and network control.
The project is part of a larger collaborative project, the PEPR 5G and next generation networks, involving a large number of French research laboratories working on networking. This will be a unique opportunity to collaborate within the networking community.
A PhD grant is funded as part of the global project.
The internship can thus be followed by a PhD for a motivated student.
Who?
• Name: Walid Dabbous, Damien Saucez
• Mail: Walid.Dabbous@inria.fr, Damien.Saucez@inria.fr
• Web pages:
• https://team.inria.fr/diana/walid-dabbous/
• https://team.inria.fr/diana/damien-saucez/
Where?
• Place of the project: Diana project-team, Inria centre at Universite Cote d'Azur
• Address: 2004, route des lucioles, 06902 Sophia Antipolis, France
• Team: Diana
• Web page: http://team.inria.fr/diana/
Pre-requisites if any: good knowledge of 5G protocols is a plus; system experience with Linux is a plus.
Description:
LiDAR scan their environment to make high resolution maps of their environment. More and more applications use LiDARs to have a precise model of their environment. Raw data taken with LiDAR must be processed by 3D modellers that are computational intensive. In many situations, it is thus preferable to process the raw data away from the device with the LiDAR itself (e.g., a drone, a car...) as it might be too expensive to equip with appropriate compute resources, compromise its battery lifetime, or cause overheating. A solution is to upload the raw data from the LiDAR to the cloud. In practice, 5G is an excellent candidate for this but the uplink bandwidth required for real applications reaches the limits of 5G.
In this Internship (possibly followed by a post-graduate stay as PhD student or engineer), we will push the performance limits of our devices to see how far can we go. The student will operate real 5G gNB and 5G industrial UEs based on OpenAirInterface and tune them to improve their performance.
Our LiDARs have to work at 10Hz, the target is then to reach an upload of 1,75 M of floats every 100 ms, which translates in uplinks of ~150Mbps with bounded delay, which hasn't been reach yet with strictly open source software. Multiple LiDARs emit at the same time.
To optimise communications, LiDAR data will be used to construct real-time 3D models of the environment and configure Reconfigurable Intelligent Surfaces (RIS) to help improving communication efficiency. Intelligent surface can be seen as a beam former that can be used to "re-beam" the signal on the fly.
The student is expected to:
1. Theoretically provide the appropriate 5G/MIMO profiles that guarantee the required performance
2. Build a benchmarking framework to automatically determine software and hardware bottlenecks for OAI RAN
3. Setup experiments that carry LiDAR flows over 5G
Useful Information/Bibliography:
[OAI] https://openairinterface.org/
Name: Chuan Xu, Giovanni Neglia
Mail: chuan.xu@inria.fr, giovanni.neglia@inria.fr
Webpage: https://sites.google.com/view/chuanxu,
http://www-sop.inria.fr/members/Giovanni.Neglia/
LOCATION
Inria Sophia-Antipolis Méditerranée
Address: 2004 route des Lucioles, 06902 Sophia Antipolis
Team: COATI
Webpage: https://team.inria.fr/coati/
DESCRIPTION
Federated Learning (FL) empowers a multitude of IoT devices, including mobile phones and sensors, to collaboratively train a global machine learning model while retaining their data locally [1,2]. A prominent example of FL in action is Google's Gboard, which uses it to predict subsequent user inputs on smartphones [3].
Two primary challenges arise during the training phase of FL [4]:
Data Privacy: Ensuring user data remains confidential.
Security Against Malicious Participants: Ensuring the learning process isn't derailed by harmful actors.
Addressing the issue of privacy, Differential Privacy (DP) algorithms have been developed. These algorithms work by adding noise to transmitted messages, ensuring that minor alterations in a user's training sample won't be discernible to potential adversaries [5,6]. In terms of resilience against malicious actors, Byzantine resilience algorithms are employed to sift out potentially harmful user updates [7,8,9,12].
However, a challenge arises when combining these techniques. Recent studies indicate that when these methods are directly combined, the resulting algorithm's reliability becomes heavily contingent on the number of parameters in the ML model. This makes training expansive models nearly unfeasible [10]. Hence, there's a pressing need for innovative methods that can seamlessly integrate DP and Byzantine resilience [11].
For this internship, we expect the student to:
Familiarize himself/herself with the intricacies of Federated Learning and comprehend the theoretical challenges highlighted in [10].
Implement the method proposed in [11] using PyTorch.
Evaluate its effectiveness in maintaining privacy and its robustness against malicious threats within the FL framework.
PREREQUISITES
We are looking for a candidate with coding experience in Python and good analytical skills.
REFERENCES
[1] McMahan et al, Communication-Efficient Learning of Deep Networks from Decentralized Data, AISTATS 2017, pages 1273-1282
[2] Tian Li et al, Federated learning: Challenges, methods, and future directions. IEEE Signal Processing Magazine, pages 50-60, 2020
[3] Hard, Andrew et al, Federated Learning for Mobile Keyboard Prediction. arxiv: 1811.03604, 2019
[4] Kairouz et al, Advances and Open Problems in Federated Learning
[5] McMahan et al, Learning differentially private recurrent language model, ICLR 2018
[6] Bellet et al, Personalized and Private Peer-to-Peer Machine Learning, AISTATS 2018
[7] Yin et al, Byzantine-Robust Distributed Learning: Towards Optimal Statistical Rates
[8] Krishna Pillutla et al, Robust Aggregation for Federated Learning
[9] Blanchard et al, Machine Learning with Adversaries: Byzantine Tolerant Gradient Descent, NIPS 2017
[10] Guerraoui et al, Differential Privacy and Byzantine Resilience in SGD: Do They Add Up?, PODC 2021
[11] Guerraoui et al, Combining Differential Privacy and Byzantine Resilience in Distributed SGD
[12] Guerraoui et al, Byzantine Machine Learning: A Primer. ACM Comput. Surv., August 2023
-
Advisors: APARICIO PARDO Ramon, URVOY-KELLER Guillaume
Mails: raparicio@i3s.unice.fr, guillaume.urvoy-keller@univ-cotedazur.fr
Web pages: http://www.i3s.unice.fr/~raparicio/, https://webusers.i3s.unice.fr/~urvoy/
Place of the internship:
Address: I3S: Laboratoire d'Informatique, Signaux et Systèmes de Sophia Antipolis
2000, route des Lucioles - Les Algorithmes - bât. Euclide B, 06900 Sophia Antipolis
Team: Signet
Web page: http://signet.i3s.unice.fr
Description:
In the recent years, Data-Centers Micro-Grids (DCMGs) are gaining attraction as a promising
solution to reduce the carbon fingerprint of the Content Delivery Networks (CDN).
Microgrids are reduced versions of electrical grids (electricity delivery networks) connected
to the public main electrical grid by a single point. This connection acts as a switch that
allows the microgrid to be “disconnected” from the public grid. When it is disconnected, we
say that the microgrid functions autonomously in “island mode.” A microgrid consists of
three elements: (i) a local energy production facility with a low carbon fingerprint (e.g.,
photovoltaic panels, wind turbines, …); (ii) a storage system (e.g., batteries, water reserve
for pumped storage, …); (iii) an intelligent management system to ensure the constant
balance between electricity production and demand. The increasing interest of CDNs
operators, e.g., Akamai, for this solution [1] is driven by three reasons [2]: (1) new
regulations from governments worried by the datacenters impact on satisfaction of the
national climate goals, (2) operator interest in becoming less reliant on the main electrical
grid (e.g., in case of major power outage); and (3) even a model business shift to pass from
intensive power consumer to an eventual power producer, i.e., to also sell the electrical
production surplus. The aforementioned three factors could contribute to support future
datacenters' growth, which is not evident in the current context of shortage of energy.
Unfortunately, the renewable nature of the energy sources in microgrids introduces
intermittence and uncertainties in the power generation. Moreover, the real-time workload,
i.e., the datacenter power demand, is also itself uncertain. This double uncertainty in energy
production and consumption “impacts on the operational efficiency of DCMG, especially
when it is in the island mode” [3]. There, the management of Data-Centers Micro-Grids for
minimizing carbon fingerprint whereas the workload is served supposes a challenging
optimization problem where the optimal policy could be found suing classical optimization
theory as Linear Programming [3].
In this internship, we aim to design control algorithm targeting a reduction on the carbon
fingerprint of the Data-Centers Micro-Grids by reducing the QoS provided by the applications
hosted in the datecenters. The objective is to adapt the QoS to the availability of green
energy available since small degradations in QoS, almost imperceptible for end-users, can
introduce significant power consumptions reductions.
Expected skills
Languages:
- Python language (absolutely)
- Optimization solvers, as IBM Cplex
- Deep Learning libraries (like TensorFlow [6], Keras, rllab, OpenAI Gym) (recommended)
Theory:
- Machine Learning, Data Science, particularly Neural Networks theory (recommended)
- Classical optimization theory (Linear Programming, Dual Optimization, Gradient
Optimization, Combinatorial Optimization, Graph Theory) (recommended)
Technology:
- Computer networking (recommended)
Useful Information/Bibliography:
[1] https://www.akamai.com/newsroom/press-release/akamai-to-power-dallas-data-
center-with-sustainable-energy
[2] https://www.energytech.com/distributed-energy/article/21250343/why-do-data-
centers-need-their-own-microgrid
[3] Lian, Yicheng, et al. "Robust multi-objective optimization for islanded data center
microgrid operations." Applied Energy 330 (2023): 120344:
https://www.sciencedirect.com/science/article/pii/S0306261922016014
Advisor: APARICIO PARDO Ramon
Mail: raparicio@i3s.unice.fr
Web page: http://www.i3s.unice.fr/~raparicio/
Place of the internship:
Address: I3S: Laboratoire d'Informatique, Signaux et Systèmes de Sophia Antipolis
2000, route des Lucioles - Les Algorithmes - bât. Euclide B, 06900 Sophia Antipolis
Team: Signet
Web page: http://signet.i3s.unice.fr
Description:
In the long term, Quantum Communications promise to connect Quantum Processors
placed at remote locations, giving rise to a so-called Quantum Cloud able to perform some
specific computation tasks in much shorter processing times than nowadays. In the short
term, Quantum Communications are applied in tasks such as cryptography key distribution
or clock synchronization [1]. In both cases, the basic “operation” necessary to carry out is to
set up and to keep a quantum entanglement between the communication end nodes.
Unfortunately, quantum entanglement is a probabilistic process strongly dependent on the
features of involved devices (optical fibers, lasers, quantum memories, …), whose outcome
cannot be easily estimated.
The management decisions (i.e. the control policy) to set up and to keep the
entanglement as long as possible with the highest quality constitutes a stochastic control
problem. This Entanglement Management Problem can be modelled as Markov Decision
Process (MDP) and solved under the Reinforcement Learning (RL) framework (a form of
Machine Learning) [2-3].
In this internship, we aim to find optimal control policies for setting up and keeping the
entanglement by making use of Reinforcement Learning algorithms.
To do that, we will follow the next steps:
1. Model the Quantum Entanglement Management Problem as a Markov Decision
Process.
2. Implement this model as RL environment by making use of OpenAI Gym [4], a
Python popular reinforcement learning toolkit.
3. Solve the problem using the RL algorithms implemented on the library OpenAI
Baselines [5]
4. Discuss the results by the RL algorithms by comparing them with some known
optimality bounds.
Expected skills
Languages:
-Python language (absolutely)
- Deep Learning libraries (like TensorFlow [6], Keras, rllab, OpenAI Gym) (recommended)
Theory:
- Machine Learning, Data Science, particularly Neural Networks theory (recommended)
- Classical optimization theory (Linear Programming, Dual Optimization, Gradient
Optimization, Combinatorial Optimization, Graph Theory) (recommended)
Technology:
- Computer networking (recommended)
- Quantum computing and networking (not necessary but convenient)
Useful Information/Bibliography:
[1] "Quantum Networks: From a Physics Experiment to a Quantum Network System" with Stephanie Wehner,
[Online]. Available: https://www.youtube.com/watch?v=yD193ZPjMFE
[2] S. Khatri, “Towards a General Framework for Practical Quantum Network Protocols,” LSU Doctoral
Dissertations, Mar. 2021, [Online]. Available:
https://digitalcommons.lsu.edu/gradschool_dissertations/5456
[3] S. Khatri, “Policies for elementary links in a quantum network,” Quantum, vol. 5, p. 537, Sep. 2021, doi:
10.22331/q-2021-09-07-537.
[4] “Gym: Gym toolkit for creating reinforcement learning environments.” [Online]. Available:
https://gym.openai.com
[5] “Openai baselines: high-quality implementations of reinforcement learn- ing algorithms.” [Online].
Available: https://github.com/openai/baselines
Who? Name: Giovanni Neglia, Alain Jean-Marie, Angelo Rodio
Mail: firstname.familyname@inria.fr
Web page: http://www-sop.inria.fr/members/Giovanni.Neglia/
Where?
Place of the project: Inria
Address: 2004 route des Lucioles, 06902 Sophia Antipolis
Team: NEO team
Web page: https://team.inria.fr/neo/
Pre-requisites if any: The ideal candidate should like math and analytical reasoning and have strong programming skills. A background on optimization or machine learning would be a plus.
Description:
The increasing size of data generated by smartphones and IoT devices motivated the development of Federated Learning (FL) [1,2], a framework for on-device collaborative training of Machine Learning (ML) models. FL algorithms like FedAvg [3] allow clients to train a global ML model without sharing their personal data; FL reduces data collection costs and protects clients' data privacy. However, as edge devices grow in computational power, communication over wireless networks has emerged as a bottleneck in FL, and has posed new challenges such as degradation in training performance due to transmission losses [4,5]. To address this issue, conventional solutions enhance network robustness via resource allocation [5], retransmission [6], and error correction [7], but extend training time and energy consumption. Nonetheless, the stochastic nature of gradient methods offers inherent resilience to limited errors, enabling convergence despite packet losses [8,9].
Our research project aims to investigate novel algorithmic approaches to improve communication efficiency in FL training. This includes dealing with packet losses [8-10], using model compression and sparsification techniques [11], leveraging the existing model quantization in digital transmissions [12], and exploring the use of previous gradient information to expedite model convergence [13,14].
This research topic can lead to a PhD position. We are actively looking for students with a strong motivation to pursue a research career.
Useful Information/Bibliography:
[1] Li, Tian, Anit Kumar Sahu, Ameet Talwalkar, and Virginia Smith. 2020. “Federated Learning: Challenges, Methods, and Future Directions.” IEEE Signal Processing Magazine 37 (3): 50–60.
[2] Kairouz, Peter, H. Brendan McMahan, Brendan Avent, Aurélien Bellet, Mehdi Bennis, et al. 2021. “Advances and Open Problems in Federated Learning.” Foundations and Trends® in Machine Learning 14 (1–2): 1–210.
[3] McMahan, Brendan, Eider Moore, Daniel Ramage, Seth Hampson, and Blaise Aguera y Arcas. 2017. “Communication-Efficient Learning of Deep Networks from Decentralized Data.” In Proceedings of the 20th International Conference on Artificial Intelligence and Statistics, 1273–82. PMLR.
[4] Yang, Howard H., Zuozhu Liu, Tony Q. S. Quek, and H. Vincent Poor. 2020. “Scheduling Policies for Federated Learning in Wireless Networks.” IEEE Transactions on Communications 68(1): 317–33.
[5] Chen, Mingzhe, Zhaohui Yang, Walid Saad, Changchuan Yin, H. Vincent Poor, and Shuguang Cui. 2021. “A Joint Learning and Communications Framework for Federated Learning Over Wireless Networks.” IEEE Transactions on Wireless Communications 20 (1): 269–83.
[6] Wen, Dingzhu, Xiaoyang Li, Qunsong Zeng, Jinke Ren, and Kaibin Huang. 2019. “An Overview of Data-Importance Aware Radio Resource Management for Edge Machine Learning.” Journal of Communications and Information Networks 4 (4): 1–14.
[7] Su, Xiaoxin, Yipeng Zhou, Laizhong Cui, and Jiangchuan Liu. 2023. “On Model Transmission Strategies in Federated Learning With Lossy Communications.” IEEE Transactions on Parallel and Distributed Systems 34 (4): 1173–85.
[8] Salehi, Mohammad, and Ekram Hossain. 2021. “Federated Learning in Unreliable and Resource-Constrained Cellular Wireless Networks.” IEEE Transactions on Communications 69 (8): 5136–51.
[9] Rodio, Angelo, Giovanni Neglia, Fabio Busacca, Stefano Mangione, Sergio Palazzo, et al. 2023. "Federated Learning with Packet Losses." To appear in the IEEE International Symposium On Wireless Personal Multimedia Communications.
[10] Ye, Hao, Le Liang, and Geoffrey Ye Li. 2022. “Decentralized Federated Learning With Unreliable Communications.” IEEE Journal of Selected Topics in Signal Processing 16 (3): 487–500.
[11] Wang, Bin, Jun Fang, Hongbin Li, and Bing Zeng. 2023. “Communication-Efficient Federated Learning: A Variance-Reduced Stochastic Approach With Adaptive Sparsification.” IEEE Transactions on Signal Processing, 1–15.
[12] Zheng, Sihui, Cong Shen, and Xiang Chen. 2021. “Design and Analysis of Uplink and Downlink Communications for Federated Learning.” IEEE Journal on Selected Areas in Communications 39 (7): 2150–67.
[13] Gu, Xinran, Kaixuan Huang, Jingzhao Zhang, and Longbo Huang. 2021. “Fast Federated Learning in the Presence of Arbitrary Device Unavailability.” In Advances in Neural Information Processing Systems, 34:12052–64.
[14] Jhunjhunwala, Divyansh, Pranay Sharma, Aushim Nagarkatti, and Gauri Joshi. 2022. “Fedvarp: Tackling the Variance Due to Partial Client Participation in Federated Learning.” In Proceedings of the Thirty-Eighth Conference on Uncertainty in Artificial Intelligence, 906–16. PMLR.
Who?
Name: Françoise Baude & Fabrice Huet
Mail: francoise.baude@univ-cotedazur.fr, fabrice.huet@univ-cotedazur.fr
Telephone:
Web page: https://scale.i3s.unice.fr/
Where?
Place of the project: I3S laboratory
Address:
Team: Scale
Web page:https://scale.i3s.unice.fr/
Distributed event queues have become a central component in constructing
large-scale and real-time cloud applications. They are currently
employed in various latency-sensitive cloud applications, such as
recording and analyzing web accesses for recommendations and ad
placement, health care monitoring, fraud detection, smart grids, and
intelligent transportation.
A distributed event queue comprises several partitions or sub-queues
deployed across a cluster of servers. Applications (Event Consumers)
that pull and process events from distributed queues are
latency-sensitive. They necessitate a high percentile of events to be
processed within a desired latency. Overprovisioning resources to meet
this latency requirement is suboptimal, as it incurs substantial
monetary costs for the service provider. Therefore, designing solutions
for resource-efficient and latency-aware event consumers from
distributed event queues is crucial. Such an architecture should
dynamically provision and deprovision resources (event consumer
replicas) to minimize resource usage while ensuring the required service
level agreement (SLA).
To achieve this objective, we have framed the problem of autoscaling
event consumers from distributed event queues to meet a desired latency
as a bin pack problem. This bin pack problem is dependent on the arrival
rate of events into queues, the number of events in the queues backlog,
and the maximum consumption rate of the event consumers. We have
validated our approach through extensive experiments where a service
dynamically scales based on the input load.
The aim of this internship is to expand this work to multi-service
scenarios. Specifically, we aim to investigate how to dynamically scale
a Directed Acyclic Graph (DAG) of microservices while maintaining the
SLA. In this context, any scaling action on an upstream microservice
will have repercussions downstream. The intern will be responsible for
the following tasks:
- Conduct a comprehensive review of existing literature on scaling
DAGs of services, focusing on comparing approaches for RPC based against
event queue based interconnections between (micro)services in such DAGs.
- Propose modifications to our current algorithms to incorporate the
complexities of a DAG.
- Execute thorough experiments to validate the proposed changes.
Pre-requisites if any:
- Docker
- Java
References:
Mazen Ezzeddine, Gael Migliorini, Françoise Baude, Fabrice Huet. Cost-Efficient and Latency-Aware Event Consuming in Workload-Skewed Distributed Event Queues. 6th International Conference on Cloud and Big Data Computing (ICCBDC’2022)
G. Shapira , T. Palino, R. Sivaram and K. Petty. Kafka: The Definitive Guide: Real-Time Data and Stream Processing at Scale, second edition, O’Reilly Media, Inc., 2021.
Y.Ajiro, A.Tanaka. Improving packing algorithms for server consolidation. In int. CMG Conference,Vol. 253, 2007.
Name: Christelle Caillouet
Mail: christelle.caillouet@univ-cotedazur.fr
Web page: http://www-sop.inria.fr/members/Christelle.Molle-Caillouet/
Place of the project: Inria
Address: 2004 Route des Lucioles, 06902 Sophia Antipolis
Team: Coati
Web page: https://team.inria.fr/coati/
Pre-requisites if any: Algorithms for Telecommunications
Strong programming skills: python, scripting, Java, C, etc.
Description:
The goal of this internship is to optimize the mapping of multi-actors service requirements to available resources in a multi-party global service architecture while avoiding the exchange of confidential information [4]. To do so, we propose to study the impact of exchanging only abstractions of each domain on the service establishment. A trade-off must then be found between the precision of the information exchanged and the respect of the confidentiality property. The information exchanged must be sufficiently precise to ensure the respect of the service constraints while preventing the deduction of the topology or the performance of a domain. Network abstractions have been studied in the past to reduce the complexity of routing optimization [1][2], focusing on the link resources of the network. This is clearly not addressing slice placement problems that need also to consider node resources.
We will investigate how to abstract the network and evaluate the consequence of exchanging network abstractions for the orchestration of services that needs to guaranty E2E QoE, energy consumption, and security [3]. To do so, we will investigate several solution to abstract a topology using graph theory for example. The evaluation of the abstraction performance will be done by estimating its impact on virtual service placement problems that can be formulated as a maximum utility optimization problem from a multi-user multi‐party perspective.
The project is part of a national collaborative project, the PEPR NF (Network of the Future). The internship can be followed by a secured PhD grant funded by the project for motivated student.
Useful Information/Bibliography:
[1] Piet Van Mieghem. "Topology information condensation in hierarchical networks." Computer networks 31.20 (1999) : 2115-2137.
[2] M. Scharf, T. Voith, M. Stein and V. Hilt, "ATLAS : Accurate Topology Level-of-Detail Abstraction System," 2014 IEEE Network Operations and Management Symposium (NOMS), Krakow, Poland, 2014, pp. 1-5, doi : 10.1109/NOMS.2014.6838357
[3] Ahmad, A., Schultz, A., Lee, B., & Fonseca, P. (2023). An Extensible Orchestration and Protection Framework for Confidential Cloud Computing. In 17th USENIX Symposium on Operating Systems Design and Implementation (OSDI 23) (pp. 173-191).
[4] De Sousa, N. F. S., Perez, D. A. L., Rosa, R. V., Santos, M. A., & Rothenberg, C. E. (2019). Network service orchestration: A survey. Computer Communications, 142, 69-94.
Who?
• Name: Manel Khelifi, Damien Saucez, Walid Dabbous
• Mail: manel.khelifi@inria.fr, Damien.Saucez@inria.fr, walid.dabbous@inria.fr
• Web pages: https://team.inria.fr/diana
Where?
• Place of the project: Diana project-team, Inria centre at Universite Cote d'Azur
• Address: 2004, route des lucioles, 06902 Sophia Antipolis, France
• Team: Diana
• Web page: http://team.inria.fr/diana/
Pre-requisites if any: Matlab, C/C++ programming experience, ns-3
Description:
The development and use of metasurfaces, also known as reconfigurable intelligent surfaces (RIS), is providing a means of creating future large 6G communications networks and reducing the cost of multiple antennas or high power consumption. These metasurfaces contain small scattering meta-atoms that disperse the incoming wave with a controllable delay, resulting in a controllable phase shift. In order to explore this area, the main focus of this internship is to:
- Review and study the research literature on user equipment (UE)-RIS-base station (BS) channel models to understand the key parameters and characteristics.
- Define the communication techniques based on the RIS channel model.
- Use the INRIA channel model already developed for the RIS, which captures both the signal propagation and reflection behavior, to implement the RIS-based communication model in NS-3. This involves writing custom C++ code within the NS-3 framework to implement the RIS channel model and defining methods to handle the interaction of the RIS with the BS and UE, as well as modelling the signal reflection, absorption and propagation across the radio spectrum.
- Validate and verify the implementation by performing simulation of different scenarios and comparing the results with expected results and/or with analytical models or simulation results from other platforms.
Useful Information/Bibliography:
[1] https://www.etsi.org/technologies/reconfigurable-intelligent-surfaces
[2] Y. Cao, T. Lv and W. Ni, "Intelligent Reflecting Surface Aided Multi-User mmWave Communications for Coverage Enhancement," 2020 IEEE 31st Annual International Symposium on Personal, Indoor and Mobile Radio Communications, London, UK, 2020, pp. 1-6, doi: 10.1109/PIMRC48278.2020.9217160.
[3] R. Sun, W. Wang, L. Chen, G. Wei and W. Zhang, "Diagnosis of Intelligent Reflecting Surface in Millimeter-Wave Communication Systems," in IEEE Transactions on Wireless Communications, vol. 21, no. 6, pp. 3921-3934, June 2022, doi: 10.1109/TWC.2021.3125734.
1 Project Members
Killian Castillon <castillon@i3s.unice.fr> - (Scale/SigNet),
Dino Lopez Pacheco <dino.lopez@univ-cotedazur.fr> - (SigNet),
Fabrice Huet <Fabrice.HUET@univ-cotedazur.fr> - (Scale)
2 Introduction
To provide new network services (e.g. routing, firewalling, etc.) and keep high performances, hardware devices need to be deployed. Indeed, the ASICs embedded into the hardware can fulfill all those requirements. However, hardware devices are difficult to update, which is a critical operation in case of bugs or protocols/services modifications.
Virtual network devices, or Virtual Network Functions (VNFs) as they are called, appear as a solution to the rigidity of hardware devices. Indeed, VNFs are deployed in general purpose servers and updating a VNF implies to only update a piece of software by mean of traditional software update techniques. Nevertheless, VNFs very hardly follows the line rate speed in high performance network, such as Data Centers.
The low performance observed in VNFs is the results of the complex path followed by network packets when they reach a server: in a standard system, packets are received by the port of a NIC, then they must go through the entire protocol stack implemented at the server’s kernel (potentially multiple times when tunneling is employed) before reaching the VNF application [1,2].
To increase the performance of VNFs, two main solutions exist today: DPDK and AF_XDP. Both rely on the idea to take the packet as soon as possible as it enters the NIC and hide it to the server’s kernel, so a direct path exists between the NIC and the VNF. The main difference between DPDK and AF_XDP is found on the interaction level between those solutions and the system. While DPDK can be considered as a parallel system running along the one of the main servers (so the server is completely unaware of the NIC and the path to the VNFs), XDP add a very light datapath very close to the NIC, which is able to forward packet to the fast XDP path or to the traditional protocol stack of the server.
3 OVS and OVS AF_XDP
One VNF frequently deployed in Data Center with Linux-based systems is OVS (Open vSwitch) [3]. OVS is an OpenFlow-based SDN switch able to provide layer 2 retransmission by default, but also, many other services when configured to work with an OpenFlow controller.
OVS are faster devices compared to the legacy software switches available in Linux (i.e. Linux bridges) and more powerful (Linux bridges do not support the OpenFlow protocol). However, despite the multiple optimizations introduced at the core of OVS, it hardly reaches the speed line rate of 10 giga Ethernet cards or higher.
Thus, OVS decided to add a support for one of the fast packet processing solutions. On the one hand, since DPDK completely steals the packet from the main system, it is mandatory for DPDK-based applications to rewrite any network tool which may be needed, but long time available in the Linux systems. On the other hand, XDP being natively supported by the Linux kernel and still having access to the systems tools, XDP-based applications need less code rewrite and less maintenance efforts. Hence, today, an experimental implementation of OVS- XDP is available [4].
4 Project objectives
At the SigNet and Scale teams, we started studying the benefits of XDP on the network performance. This is why we aim now at exploring and understanding the performance of the OVS-XDP solution. In this project, the student is expected to:
• Deploy OVS AF_XDP and carry out an experimental study to determine the maturity of this solution.
o Investigate the root of possible bugs that may appear (and solve it if possible)
• Analyze the benefits of OVS AF_XDP over the legacy OVS.
• Deploy OVS AF_XDP to interconnect Docker containers.
• Explore the integration of OVS AF_XDP into Docker Swarm or Kubernetes
5 Required skills
• Good knowledge of C programming
• Linux administration
• Good knowledge of networking (both layer 2 and layer 3)
• Docker Swarm and Kubernetes would be a plus
Bibliography
[1] Zhuo D et al., “Slim: OS Kernel Support for a Low-Overhead Container Overlay Network”. 16th USENIX Symposium on Networked Systems Design and Implementation (NSDI 19), Boston, MA, USA.
[2] Jiaxin Lei et al. "Tackling parallelization challenges of kernel network stack for container overlay networks”. In Proceedings of the 11th USENIX Conference on Hot Topics in Cloud Computing (HotCloud'19). USENIX Association, USA, 9.
[3] Ben Pfaff et al. “The design and implementation of open vSwitch”. In Proceedings of the 12th USENIX Conference on Networked Systems Design and Implementation (NSDI'15). USENIX Association, USA, 117–130
[4] William Tu et al. “Revisiting the open vSwitch dataplane ten years later”. In Proceedings of the 2021 ACM SIGCOMM 2021 Conference (SIGCOMM '21). Association for Computing Machinery, New York, NY, USA, 245–257. https://doi.org/10.1145/3452296.3472914
Who
Name: Peraldi Marie-agnes / Liquori Luigi
Mail: map@unice.fr Luigi Liquori@inria.fr
Telephone:
Web page: https://www.i3s.unice.fr/~map/ https://luigiliquori.wixsite.com/atinria
Where?
Place of the project: INRIA – Kairos team
Address: INRIA – Sophia Méditerranée
Team: Kairos
Web page: https://team.inria.fr/kairos/
What? integration and exploitation of sustainability in standards.
Pre-requisites if any:
Interest for
Modelling distributed systems
Concepts of discrete time vs simulation time
Java/eclipse knowledge is welcome
Keywords : IoT standards, sustainability, evaluation, deployment, oneM2m, Omnet++ simulation
Detailed description:
Context
Kairos team is highly involved in ETSI standardization bodies around the topic of IoT systems modeling and simulation.
oneM2M, the global standard initiative for Machine to Machine (M2M) communications and the IoT, is now mature and multiple deployments exist all over the world at both experimental and operational levels.
The internship is defined in the context of an ETSI Testing Task Force (TTF) on Performance Evaluation and analysis of oneM2M Planning and deployment. This task force started in November for 2 years. Kairos participated in this task force.
The objective is to define a data model and associated behavioral model to characterize an IoT distributed application, the targeted platform, and the deployment scenarios of the application on the platform.
The Kairos team is largely involved in temporal models for highly constrained cyber-physical systems such as in avionics or in the automotive domain. The DSL (Domain Specific Language) approach (Gemoc) using formal methods is one of the approaches developed in the team to specify and verify the behavior of these critical systems.
We want to experiment this DSL approach to characterize IoT applications, platforms and their deployment.
To do so, the student will have to draw inspiration from existing works in the field and propose at least for distributed IoT applications a temporal DSL allowing to make an executable model for the future. The execution target is the Omnet++ discrete simulator.
Objective of internship
The oneM2M standard identifies and handle different aspects of the sustainability from the point of view of the energy performances related to IoT devices, networks and data management from the deployment and operational point of view.
In this project we focus on the specific concerns of sustainability in IoT systems that imply different aspects such as the impact on carbon footprint, the longevity, performance evaluation, the energy consumption… of the deployments related to the standard and their implementation.
In this project we want to :
Identify and define a set of classified requirements related to the design, the deployment and the evaluation of multi-criteria sustainability of IoT systems,
Evaluate the coverage of the oneM2M standard w.r.t. these requirements
Provide a PoC (proof of concept) based on the simulation tools OMNeT++ for a characterization of sustainability as a key performance Index in a oneM2M system evaluation.
Thus, the student will :
review the literature on sustainability in IoT starting with these papers and exploring others to provide a state of the art analysis. [1] [2],
identify the requirements and the associated artifacts useful for the specification of sustainability criteria for the IoT domain,
Identify the KPIs (Key Performance Indexes) that can be deduced for an analysis of sustainability in a IoT deployment
If possible but not mandatory – Based on an existing DSL for oneM2M , extend this DSL with these artifacts based on model driven engineering technics [4]
NB- for the last point, preliminary works have already been conducted by a student last year and a preliminary POC (proof of concept) has been developed.
References:
set of bibliographical references (article, books, white papers, etc) to be read by the student before starting to work on this subject
[1] oneM2M ETSI - How oneM2M is enabling more sustainable IoT deployments- oneM2M white paper, sept 2022, https://www.onem2m.org/images/images/files/oneM2M-SSC-White-Paper-2nd-edition.pdf
[2] Building Information Modeling and Internet of Things integration for smart and sustainable environments: A review, Journal of Cleaner Production, Volume 312, 20 August 2021, 127716 https://www.sciencedirect.com/science/article/abs/pii/S095965262101934X
[3] TS 103716: ETSI SmartM2M; oneM2M Discovery and Query solution(s) simulation and performance evaluation, \url{https://hal.inria.fr/hal-03261059}
[4] Site web gemoc studio https://projects.eclipse.org/projects/modeling.gemoc