Tuesday, May 18, 2021
Do network engineers dream of full automation ?
17:00-18:30 (CEST UTC/GMT +2)
Nikos Anerousis, IBM, USA
Christian Jacquenet, Orange Labs, France
Kireeti Kompella, Juniper Networks, USA
Marina Thottan, Nokia - Bell Labs, USA
Panel Moderators: Prosper Chemouil, Cnam, France
The Internet has become the global infrastructure that supports a wide range of service offerings. As these services not only grow in variety but also in complexity, their design, delivery and operation have become a subtle yet complex alchemy that often requires various levels of expertise. Most of these services have been deployed for the past three decades primarily based on static service production procedures that are more and more exposed to the risk of erroneous configuration commands, among others.
Multi-service, multi-protocol, multi-technology convergent and dynamically-adaptive networking environments have therefore become one of the major challenges faced by service providers.
The diversity and the complexity of these services have been raising technical challenges for many years, from both design and operational perspectives, fed with the idea of programmable networking that was introduced about 3 decades ago. The emergence of Software-Defined Networking as well as Network Functions Virtualization that has often been the opportunity to make debatable promises about their so-called flexibility or their intrinsic ability to facilitate the automation of service delivery procedures.
But reality is much different.
Claimed automation is currently mostly restricted to the elaboration and the execution of configuration scripts, which reflect the application of decision-making procedures that remain manually declarative. In addition, this rather embryonic automation framework mostly deals with tasks that remain local to a device to the detriment of a global, network-wide, systemic view that would be able to guarantee the global consistency of the set of actions taken to design, deliver and operate a service.
Network automation is actually way more polymorphic.
So what would it take to make the network automation dream come true? Or do early SDN/NFV deployments already promise a nightmare?
This industry panel aims at exploring some of the challenges of network automation. Discussions between the panelists and the audience will address the following topics:
- Understand requirements. The management of service orders is hardly automated, besides the works conducted by the TM Forum and the Metro Ethernet Forum for the past couple of years. For example, Intent-Based Networking (IBN) techniques have been introduced a few years ago to facilitate the expression of requirements and constraints that will be eventually derived into a set of configuration instructions. The basic concept of IBN corresponds to the vision of “Don’t tell me what to do! Tell me what you want”. Can IBN apply to complex services like those provided by network slices? What’s the status of IBN anyway besides the academic work?
- Migrate. Introducing automation techniques in networks cannot be done overnight. In particular, legacy techniques will not be decommissioned in the blink of an eye. Which suggests a thorough elaboration of migration strategies: how should SDN/NFV computation logics behave with legacy gear? How can legacy techs affect the enforcement of such strategies? What’s the impact on resource allocation cycles, dynamic policy enforcement schemes, etc.? how long will the old world and the new world have to coexist? Will the old world eventually disappear? And if not, how does the old world limit the benefits of network automation?
- Set the controls for the heart of the network. How do we handle the deluge of data collected that enables the proper estimation of network conditions? How machine learning methods could help in network engineering and operations? What are the feedback mechanisms that can guarantee that what has been delivered complies with what has been negotiated, as per the outcomes of an IBN-based process? To what extent, Artificial Intelligence techniques are mature enough to address real situations? What if there’s some discrepancy between what’s delivered and what was expected? How long will it take before the network automation system and its computation logic react/adjust? And at what cost (performance, scalability, robustness)? What about the DevOps approach? Did it really improve the automation of network operation yet? What can DC providers’ and network operators’ communities learn from each other? How management frameworks like Kubernetes foster the automation of network operation? What about new approaches like Weave Works’ GitOps that claim to simplify Kubernetes management, albeit the deployment of the latter remains embryonic?
- Humans matter more than ever. What’s the cost of network automation on organizations? How to keep control of the automated delivery procedure at every stage of the process, from the dynamic, IBN-based, service parameter negotiation framework to the allocation of properly configured resources? Is automation the tolling bell of network engineer jobs?
Nikos Anerousis currently leads the network project services practice at IBM, focusing on implementations of software-defined networking, private 5G and Edge computing for IBM's enterprise clients. He received the diploma in EECS from the National Technical University of Athens in 1990 and the M.S. and Ph.D. degrees in Electrical Engineering from Columbia University in 1991 and 1996, respectively. From 1996 to 1999 he was with AT&T Bell Laboratories and AT&T Labs-Research, conducting research on network architectures and management of the early Internet, including early work on network programmability. From 1999 to 2003 he was the CTO of Voicemate, a venture-funded company that pioneered mobile voice authoring technology. From 2003 to 2016 he was with IBM Research in various staff and management positions working on distributed middleware, service management, analytics, and automation. He has a long history of contributions in network and services management and is a frequent participant in the IM and NOMS conferences.
Christian Jacquenet graduated from the Ecole Nationale Supérieure de Physique de Marseille, a French school of engineers. He joined Orange in 1989, and he’s currently the Referent Expert of the “Networks of the Future” Orange Expert community. Until recently, he was the Director of the Strategic Program Office for advanced IP networking within Orange Labs. He is also the head of Orange’s IPv6 Program that aims at defining and driving the enforcement of the Group’s IPv6 strategy and which yielded the deployment of IPv6 networks and services in most European and African Orange affiliates since 2010. He leads development activities in the areas of network automation (including SDN, automated service delivery procedures combined with Artificial Intelligence techniques, intent-based networking), and IP networking techniques. He authored and co-authored several Internet standards in the areas of dynamic routing protocols and resource allocation techniques, as well as numerous papers and books about IP multicast, traffic engineering and automated IP service delivery techniques. He also holds 20+ patents in the areas of advanced home and IP networking techniques.
Currently SVP and Engineering CTO at Juniper Networks, Kireeti Kompella was formerly CTO at Contrail Systems, and before that, CTO and Chief Architect of JunOS at Juniper Networks. Dr. Kompella has deep experience in Packet Transport, large-scale MPLS, VPNs, VPLS, and Layer 1 to Layer 3 networking. Kireeti participates in the IETF, as past chair of the CCAMP Working Group and as author of several RFCs. His responsibilities have evolved from a Network OS (Junos) to routing protocols to standards work; but now turns to network-wide considerations, including 5G networks. His focus currently is on Self-Driving Networks and the application of Machine Learning to achieve experience-first networks. Prior to Juniper, Kireeti worked on file systems at NetApp (wafl), SGI (xfs/xlv), and ACSC (acquired by Veritas). Dr. Kompella received his BS EE and MS CS at IIT, Kanpur, and his PhD in Computer Science at USC, specializing in Number Theory and cryptography.
Marina Thottan joined Bell Labs Research in 1999, and has contributed to a wide variety of projects, including Content Distribution, Routing protocols, Data over Optical networks, High Speed Router Design, Network Management and Anomaly Detection. Most recently she has been leading work on Network Orchestration, Network Slicing, and 5G Security. She is a Bell Labs Fellow and an IEEE Fellow. Marina received a Ph.D. in Electrical and Computer Engineering from Rensselaer in 2000. She has published extensively and holds several patents in the area of network technologies. She is co-author of the book “Communication Networks for Smart Grids: Making Smart Grids Real” and has also Co-edited a book on “Algorithms for Next Generation Networks”.
Prosper Chemouil is an Adjunct Senior Researcher in the "Networks & IoT Systems" (ROC) at Cedric, Cnam in Paris, France. He received the M.Sc. and Ph.D. degrees in control theory from École Centrale de Nantes, in 1976 and 1978, respectively. He is retired of Orange Labs, the R&D Unit of the Orange Group, where he was a Research Director and Chief Scientist on Future Networks. In his last position, he was the Expert Program Leader on Future Networks for the whole Orange Group. His research interests are with the design and management of new networks and technologies and their impact on network architecture, traffic engineering and control, and performance, with significant involvement in standardization at ITU-T for 25 years. For several years, he has been then focusing on cognitive management and network softwarization. In 2016, he has become the Co-Chair of the IEEE SDN Initiative and is now a member of the IEEE ComSoc Industry Communities Board and Industry Outreach Board, representing the SDN/NFV/Cloud area.
Wednesday, May 19, 2021
Research Challenges in Artificial Intelligence for Network Management (joint panel with NMRG)
11:30-13:00 (CEST UTC/GMT +2)
Panel Members: to be confirmed
Panel Moderators: Jérôme François, INRIA Nancy, France
Artificial Intelligence is one of the defining technologies of our age, enabling myriads of novel applications. Networking in general, and network management in particular, is one of the areas that stands to benefit with the promise to handle more efficiently the complexity of networks and alleviate the latency human-based operations. Over last years, multiple AI-based solutions have been proposed to address network management problems in various use cases and relying on on different technologies. The flexibility of network brought from their softwarization paves the way for a higher level of automation where the AI power will certainly play a role. On one hand, real-time analysis of various events at network, service and application level will be necessary to fastly react and automate processes. On the other hand, network with the aim of very low latency and high throughput can accelerate AI. However, in comparison with other communities like speech recognition or image processing with well-established methods, models and datasets and above clear identified problems and objectives in the area of AI, the network management community appears more scattered. Identifying the right set of challenges to be addressed in our domain is essential in order to promote an AI specifically tailored to network management.
In this panel, we will discuss the challenges that are actually specific to our domain in regards to AI and possible orientations to address them. To exploit the full A.I.'s potential, many fundamental questions still need to be answered. These include, for example:
- Is AI really suitable for networking and network management ?
- What networking and management problems AI can AI solve better than other approaches? Are there even meaningful problems to solve?
- How much about all this is hype, in which areas will we see disappointments where A.I. may fall short of expectations?
- What advances in AI are required to unlock more potential applications in network management?
- Can requirements for determinism and explainability be sufficiently reconciled with existing constraints of AI? Can AI be accepted to replace years of human expertise and finely tailored procedures in production environment?
- Is AI compatible with network constraints? Real-time constraints of network vs processing time of AI? Can the network support AI at line rate while still doing its normal operations?
- Are there barriers that will be difficult for AI to overcome?
- How to build network- or network management-specific AI ?
- How can networks better support the needs of AI applications? For example, is there a need for new management protocols or new data models to better “feed” and interact with AI systems ?
- What network and management attack vectors does the application of A.I. introduce? For example, what types of problems could poisoned or biased datasets used to train AI applications introduce? How can these new vectors be counteracted ?
- Given that processing resources in a network are limited and given that network management is inherently distributed, what are the ramifications on AI architectures? Will specialized AI architectures for NM be needed; for what applications; how will they be characterized ?
- How AI can be accommodated with "external" factors ?
- Given that many networking services are regulated, are there regulatory or legal ramifications ?
- What about AI and Internet neutrality ?
- Is our privacy endangered by AI from a networking perspective ?
- Will there still be humans in the loop and what will be their role ?
- How the network management community can progress in this area ?
- Why the transfer from academic style research to industry is so low ?
- What are the position roles for research, solution development and deployment ("industry players") ?
- Do we need standards and open source ?
These and other questions will be explored by the panelists.
Intelligent Management of Future Industrial Networks
17:00-18:30 (CEST UTC/GMT +2)
Thomas Schmidt, Hamburg University, Germany
Luca Foschin, University of Bologna, Italy
Marie-Jose Montpetit, Concordia University, Canada
Panel Moderators: Cedric Westphal, Futurewei, USA
5G networks will deploy many new network services such as ultra-reliable low latency (uRLLC), massive Machine-type communications (mMTC) and enhance Mobile Broadband (eMBB). These in turns will significantly impact how industrial networks are deployed, managed and operated.
Industrial networks have evolved from serial bus towards more complex systems, and are converging towards IT. However, they are networks with specific protocols, constraints and requirements. Currently, the main protocols are EtherNet/IP, Profinet, EtherCAT, Powerlink, Modbus-TCP and others. However, wireless networks are becoming more and more deployed, growing 30% year on year over the last few years. This evolution will only accelerate when 5G will be deployed.
Industrial networks have stringent requirements in scale, delays, security and amount of bandwidth. In particular, these networks require deterministic network to ensure low latency and reliable delivery of packets. The 802.1 TSN family of protocol, and the IETF DetNet working groups are initial answers to these requirements. More are needed, especially to provide low latency packet delivery over distributed and heterogenous networks. The bandwidth generated by industrial sensor networks will also increase exponentially.
We aim to explore how such diverse and wide range of applications can challenge how to manage future industrial networks & internetworking technologies. It involves research in protocols, architectures, security, and algorithms. We aim to discuss the challenges that new networking technologies bring to the management of industrial networks.
The panel will examine the issues underlying the management of new industrial networks and protocols. Panelists will represent the different sides of the argument. Specifically, panelists will debate whether industrial network management requires innovation beyond current protocols; whether new protocols are required; what are the new trends, challenges and shortcomings in managing industrial networks today; what is the competitive landscape; whether work at the networking layers will be sufficient to address newly emerging challenges and issues that emerging industrial applications are faced with; what are the challenges for service level management, i.e. guarantees for high-precision services; what is the impact on monitoring/measurement (are the existing techniques enough, what else is required?); does network convergence play a role here, or would we expect to manage industrial network deployments still separate from other networks?