Community Research and Development Information Service - CORDIS


CogNet Report Summary

Project ID: 671625
Funded under: H2020-EU.

Periodic Reporting for period 1 - CogNet (Building an Intelligent System of Insights and Action for 5G Network Management)

Reporting period: 2015-07-01 to 2016-06-30

Summary of the context and overall objectives of the project

The overall objective of the CogNet project is to make a major contribution towards autonomic network management through the use of network information such as state, topology, configuration and various key metric data, and relevant external data which we call situational context, and use this to proactively manage the network, maintain sufficient capacity to meet demand levels and maintain Quality of Service at the levels required by the different types of network users.

At a more detailed level, the following key objectives were identified:

1. Research and develop a system of data collection from network nodes that involves preprocessing data to allow the node classify the data it generates and identify the most important and irregular data for submission to network management while filtering routine and regular data. This is an important step in the development of scalable network management as it dramatically reduces the scale of data required to be processed centrally.

2. While working on the principles of a self organising network, research and develop, within existing policy management frameworks, a system to allow network nodes to self manage based on their available data while escalating higher importance issues to central network management.

3. Apply Machine Learning algorithms to develop a system of service demand prediction and provisioning which allows the network to resize and resource itself, using virtualisation, to serve predicted demand according to parameters such as location, time and specific service demand from specific users or user groups. This is achieved while optimising performance and use of available network and VM resources while minimising overall energy requirements and costs.

4. Apply Machine Learning algorithms to address network resilience issues. This includes using Supervised ML to identify network errors, faults or conditions such as congestion at both a network wide and a local level and automatically taking mitigating actions to minimise overall impact.

5. Use anomaly detection algorithms to identify serious security issues such as unauthorised intrusion or fraud and liaise with autonomic network management & policies to formulate and take appropriate action.

6. Develop a number of demonstrable applications using real-world data gathered via current 4G network nodes which demonstrate the core project innovations, and serve to highlight the exploitation potential of CogNet. The applications will include tests to demonstrate the potential improved performance and capacity that can be achieved by utilising the CogNet algorithms over conventional approaches used in today’s Network Management Systems.

The project goal is to develop some of the solutions that will provide this higher and more intelligent level of automated monitoring and management of networks and applications, improve operational efficiencies and facilitate the requirements of 5G. To achieve this, the project will conduct research in the areas of network data gathering, machine learning, data analytics and autonomic network management.

The research in CogNet will eventually contribute towards more reliable mobile communications which in turn will allow new applications such as autonomic driving, smart mobile medical technology, richer communications experiences, and a more energy efficient ICT infrastructure.

Work performed from the beginning of the project to the end of the period covered by the report and main results achieved so far

WorkPackage 1
Prior to official start of the project, the coordinator was responsible for the successful negotiation of the Consortium Agreement during this reporting period. The agreement for this consortium is signed and in place among all of the partners.

The first deliverable of WP1 was the Project Management Handbook which outlined the procedures roadmap for the group, in support of the Grant Agreement and Consortium Agreement. The plan details a focused and effective communications and management structure for all partners. The structure outlined delegates a certain amount of autonomy to Work Package Leaders, but all the time using the Grant Agreement, and regular meetings and calls to ensure WP leaders and all partners are supported and heard.

The second deliverable was the Data Management Plan document which outlined the processes to be used when using or processing data which may give rise to privacy issues. The plan outlined how such issues would be dealt with and included any required notifications of Data Protection Agencies in the jurisdictions concerned. This particular deliverable was particularly important for CogNet as it is certain that CogNet will deal with information that may raise such privacy issues.

WorkPackage 2
The work carried out in the first year in WP2 included:
Analysis of the use cases and scenarios of the CogNet system aiming at representative yet extensible set of deployment domains with a deployment specific set of platform services and components. This objective is directly reflected by Task 2.1 (led by IBM).

Modeling and design of the technical and business requirements for the CogNet system covering the representative set of usage scenarios and arranged in a hierarchy. This objective is directly mapped to Task 2.2 (led by TSSG).

Engineering the high-level architecture of the CogNet system as a harmonious set of services, service components and configurations that meet the requirements in all representative deployment domains, supported by a CogNet information model. This objective is directly mapped to Task 2.3 (led by IBM).

Deliverable D2.1 was finalized at the end of M5. The goal of this deliverable was to identify and define a set of use cases and scenarios and their requirements, related to the 5G network, specifically from a network management perspective. The deliverable introduced six use cases and eleven scenarios pivoting around cognitive network management, in particular leveraging machine learning for addressing the six identified CogNet challenges. Each scenario presented its definition, technical enablers and user story. The scenarios ranged from large scale events prediction to dense urban area congestion prediction.

WorkPackage 3
During the first year, one deliverable, D3.1 has been produced in M8 under the supervision of UNITN, with the contributions by all the WP3 partners and with the internal reviewing by ORANGE. 19 research papers have been published by the WP3 partners or accepted for publication, two of them winning best paper awards. This shows that the work package is progressing according to the research plan and the objectives are being achieved successfully.
After the completion of D3.1, the work package focused on implementing the first prototype of the CogNet Smart Engine, to be reported in Deliverable 3.2 (M16). This work follows two important directions. First, WP3 are implementing machine learning modules for various CogNet scenarios and use cases, as identified in WP2. This is done individually by each partner, in close collaboration with the applied work packages 4 and 5. Secondly, WP3 are designing a common SPARK-based interface for all the modules supported by individual partners. This is a collective effort by the whole WP to be then adopted by each contributor.

WorkPackage 4
Machine Learning techniques have been researched to allow scale up/down NFVi resources (e.g. virtual machines) elastically in order to achieve energy efficiency or system performance. We selected a paradigmatic problem in the context of virtual machine placement called “Noisy Neighbours” and achieved to identify the problem in real time applying off-the-shelf ML classifiers.
Research on traffic identification and classification without using packet payload (encrypted traffic & privacy laws) and applying off-the-shelf machine learning classifiers. Due to the lack of real data assets and test-beds, we have designed and deployed the first version of a realistic network laboratory in which researchers can apply machine learning techniques in a controlled way.
Research on how to ensure SLA compliancy and service availability in the context of softwarized 5G networks by applying machine learning techniques to predict SLA violation and SLOs breaches. The predictions will be used to identify, compute and enforce corrective actions.
Using social media data (e.g. Twitter, Foursquare) several machine learning techniques were identified and applied to predict large gatherings of people that might drastically affect traffic demand in a network.
Research on the application of machine learning to predict network demand according to various patterns. These predictions will be used to smart resource placement and migration to optimize network performance and user experience.
A cross-layer machine learning-based network management solution is being developed for support data rate demands by controlling the smart antennas with reflectors.
Deliverable D4.1 was produced at the end of the first period. Partners have contributed to this deliverable including in it the research activities done in the proofs of concept during this period. D4.1 includes for each proof of concept (1) a specification of the problem, (2) a description of the selected machine learning techniques and how they were applied, (3) a description of the dataset and (4) the obtained results. As proofs of concept have reached different maturity levels, the obtained results range from inexistent to preliminary.
A cross-layer machine learning-based network management solution is being developed for support data rate demands by controlling the smart antennas with reflectors.

WorkPackage 5
The work during the work package was split between the specific task work which was executed in small teams and overall work package tasks covering a specific item across the tasks.
For the overall work package, Orange and TID contributed by providing the clarification of the requirements of the different use cases as well as the integrated management-ML system specification which binds the work of the different tasks into a single system for security and resilience.
For the distributed security enablement, WIT provided the scope, background and the problem statement to be solved as well as the reference architecture model. Together with FHG and TUB, WIT provide the implementation of the ML algorithms, which was reviewed by the machine learning experts from UNITN and IBM for an efficient system. WIT and FHG provided the implementation and evaluation considerations for the distributed security solution.
For the reliability framework, FHG provided the initial problem statement and scenarios for reliability. FHG together with Orange and TUB provide the reference architecture. The machine learning algorithms were selected and their application was further specified by Fraunhofer together with IBM. The implementation considerations follow the parallel FHG testbed developments reported in WP6.
For the performance degradation correction, IBM provided the problem statement, the algorithms and the implementation considerations. In parallel, FHG provided to IBM advice on the specifics of network management as well as a set of data acquired from the testbed and UNTIN on the specific algorithms to be considered from the WP3 perspective as part of the network performance degradation detection.
An evaluation framework was designed by FHG (evaluation scope, scenarios and architectural reference model) and by TUB (design of a visualization GUI for being able to perceive in real time the situation of the network). UNITN contributed with an evaluation of the proposed framework from the perspective of the machine learning.

WorkPackage 6
The first activities in WP6 pivot on T6.1 Integration and Validation Methodology. It is led by TSSG, and mainly contributed by VIC through all the topics, TID in the standards dimension, IBM in the aspects beyond CogNet, IRT in the validation criteria and ORA in the overall review. Here, the WP6 telcos and work aimed at: defining common structures, methodologies and policies; stating the mechanism to efficiently and continuously integrate and test the different components developed in WP3, WP4 and WP5; specifying policies for the code development aiming at the integration and validation activities; identify the infrastructures for continuous integration solution; establishing a methodology and solutions for prototype evaluation; and polling the size and technologies of infrastructures considering computation needs of the Machine Learning tools and the dimension of a representative switching infrastructure to be managed and optimized. The results were compiled at D61 - Initial integration and validation plan.
Then, the T6.2 Platform Integration and Testing is hosting most of the activities carried out in WP6 in this period. It is led by VIC and with representative contributions from TSSG bringing the experience to deploy and setup all the NFV/SDN and Continuous Integration systems, IBM in the ML systems and NOK in the interfaces definition. Here, it has been deployed and setup a Continuous Integration System spanning all the systems and resources for the iterative integration of underlying components from WP3, 4, 5. The results, a Jenkins Continuous Integration System linked to a project repository in GitHub and a Common Infrastructure TestBed hosted in RackSpace are available for the consortium. Thus, sharing resources to experiment enables the adherence to a common framework utilising it in the realisation of the various demonstrators.
Moreover, a set of consolidated interfaces have been defined making uniform the information exchanged and interfaces of the main subsystems. Thus, driven by an implementation and integration perspective, the interfaces provide the way to glue and connect the different blocks involved in the common framework. In parallel, sample codes implementing the interfaces has been produced to boost the adoption across the different developments of demonstrators.
These results will be included in D62 - First release of integrated platform and performance reports.
Regarding T6.3 Integration with Complementary Technologies, some minor but mandatory activities lies, led by IBM and participated by NOK integrating Cloudband solution and FHG and TUB incorporating OpenBaton to the prototypes. This way, third party solutions, technologies and infrastructures have been integrated to create testbeds for demonstrators.
The other core activities are inside T6.4 Development and Testing of Demonstrators. It is led by VIC while all the partners participate around specific demonstrators. The demonstrators have been defined and initially implemented to land the scenarios and use cases. The list of demonstrators, that will cover the metrics that must be evaluated for the Validation of CogNet results, is:
• Media SLA, promoted by ORA and involving VIC and IRT, aims at SLA Enforcement, handling in an automated and efficient way the level of service guaranteed to a user or service by the network operator, for a Streaming Service that is running on SDN and NFV environment.
• Noisy Neighbours, promoted by NOK and supported UPM, targets optimal placement of resources in a virtualized cloud environment to reduce parameters that influence on the performance from another application or service.
• Performance Anomaly, promoted by IBM and backed by ORA and FHG, works on detecting anomalies and failures in the NFV context and their correlation to adapt the system dynamically to respond to the previously unforeseen events. This way, it pursues system stability, minimizing the impact of potential performance degradations while decreasing over-provisioning ratios.
• Traffic Classification, promoted by TID and collaborating with UPM and IBM, goes into privacy-friendly traffic classification. This way, it would enable resource management techniques to guarantee network stability and performance to satisfy SLAs and to enhance user QoE.
• Resilient Network, promoted by FHG and endorsed by TUB and IBM, covers side-effects appearing due to the other virtual networks which run in parallel. This would conduct policies to share the same infrastructure by multiple network services each with its own network functions.
• Connecting Vehicles, promoted by TSSG and fostered by VIC, analyses the way to setup a 5G infrastructure to track the human behaviours though a city. One critical factor in the connected cars scenario is network performance, where degradations could potentially affect the vehicle safety.
These results will be part of D63 - Final release of integrated platform and performance reports.
Concerning the activities of T6.5 Evaluation and Impact assessment, led by IRT, there are no representative achievements for month 12. It is too early to have clear inputs to assess the transfer of knowledge and expertise beyond CogNet. Once they come they will be included in D64 - Final evaluation and impact assessment results.

WorkPackage 7
Online Presence including Social Media
• Creation and delivery of CogNet website (
• Creation of the social media accounts: Twitter, LinkedIn, Facebook, YouTube
• Continuous update and maintenance of projects website and social accounts.
• “From the inside out” blog updated on a monthly basis.
• All the public deliverables made available on the website.

Communication and Dissemination
• Participation to conferences/events/workshops where project’s partner presented the CogNet outcomes (the complete list is provided in Annex A of the technical report)
• 22 papers published
• 6 papers accepted/submitted
• 4 invited talks
• 3 posters
• 1 national press release (Ireland)
• A number of national radio and TV interviews (Ireland)
• “1st International Workshop on Network Management, Quality of Service and Security for 5G Networks” organised and chaired by TSSG
• “First IFIP/IEEE International Workshop on Management of 5G Networks - 5GMan 2016”, organized by Orange.
• “NetCla: The ECML-PKDD 2016 Network Classification Challenge”, organized by UNITN.
• Release of D7.2
• Release of D7.3

• Identification of the different Standardisation bodies and related Working Groups which pertain to the CogNet’s research areas and topics.
• Contacts with different IRTF Working Groups in order to submit the CogNet architecture and the applicability of its use cases.
• Contacts with the SUPA WG of the IETF regarding the possible contributions of CogNet as a relevant SUPA application framework.
• Application of the outcomes of the EVE005 Group Specification of the ETSI NFV ISG, into the CogNet architecture.
• First contact with the TM Forum about the applicability of machine learning techniques for network management.
• Open Source Projects:
• Interaction with the OPNFV and ODL communities
• Direct participation in the End User Advisory Groups (EUAGs) of the OPNFV and OSM communities.
• Release of D7.6
• Release of D7.3

• Commercial exploitation plan provided by industrial partners.
• Technology transfer actions provided by academic partner and research institutions.
• Identification of Project assets.
• Release of D7.8

Progress beyond the state of the art and expected potential impact (including the socio-economic impact and the wider societal implications of the project so far)

"CogNet has achieved a number of steps beyond the state of the art over the course of the first year:

Development of an architecture that successfully applies a Machine Learning infrastructure overlay to a framework for policy based network managenment and a network infrastructire employing NFV and SDN technology.

Identification and refinement of a number of ML algorithm and theoir application to network management use cases such as Large-Scale Events, "Noisy Neighbour", Performance Degradation Detection, Geo-Analysis for Demand Prediction, Collaborative Resource Management, User QoS Determination and SLA Management.

A number of proof of concepts have been developed to varying levels of maturity, two of which have been developed to a demonstrator level in WP6.

Publication of a paper “Using Machine Learning to Detect Noisy Neighbors in 5G Networks” in July, 2016 at the EUCNC Workshop “Network Management, Quality of Service and Security for 5G Networks”. To our knowledge this is the first paper that proposes a machine learning approach to detect the Noisy Neighbours problem in a NFV scenario.

Development of an Integration environment for developing and demonstrating the project use cases. This environment brings together an Openstack based infrastructure, using OpenStack ecosystem projects, along with Telecoms Infrastructure OS projects including OpenDayLight, Open Source MANO and OpenBaton and marries it to a Machine Learning data collection and computation infrastructure to provide almost real time insights into the operations of the network.

CogNet has also engaged with external network management experts, in particular those involved in the emerging IETF SUPA standard for policy based management expression. The architecture plans to incorporate this into its overall design and would serve as an ideal test case for the SUPA standard.

The potential impact and societal benefits from this research will include development of techniques to manage the larger and more dynamic network topologies necessary in 5G, facilitate 5G applications which maintain QoS and Security levels such as Network Slicing, and lower the capital and operational costs through improved efficiencies and the use of node, link and function virtualisation. These will eventually contribute towards more reliable mobile communications which in turn will allow new applications such as autonomic driving, smart mobile medical technology, richer communications which may help reduce the need for business travel, and a more energy efficient ICT infrastructure."

Related information

Record Number: 192803 / Last updated on: 2016-12-14