Community Research and Development Information Service - CORDIS

H2020

SUPERFLUIDITY Report Summary

Project ID: 671566
Funded under: H2020-EU.2.1.1.3.

Periodic Reporting for period 1 - SUPERFLUIDITY (Superfluidity: a super-fluid, cloud-native, converged edge system)

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

Summary of the context and overall objectives of the project

The vision of the SUPERFLUIDITY project, in the framework of 5G, emerges from powerful drivers that are shaping our society.
The first driver is the increase of population and the still growing globalisation and physical and virtual mobility: more people (2 billion and a half in 1950, almost 7.5 billion today, half of them living in cities), and more interconnections among them.
The second driver is the proliferation of new or improved applications and services that need network connectivity: social networks, video (high definition), IoT (metering, smart home, connected cars), industry 4.0 (or the fourth industrial revolution, the current trend of automation and data exchange in manufacturing technologies), low latency services (games, virtual reality, autonomous vehicles), advanced services (face recognition and speech translation, cognitive and expert systems, big data exploitation).

Simply put, we have more connected devices, with each requiring higher data rates, lower latency, and ubiquitous coverage, with very high densities of users possible.

In addition, the importance of network connectivity and networked applications in our society and economy has the consequence of requiring significant improvements also in terms of: i) faster deployment of applications and services, so reducing their time to market and easing their evolution; ii) lower energy consumption; iii) enhanced security and privacy; iv) better reliability and dependability.
Furthermore, processing needs will be exacerbated in high capacity, dense networks. Current cloud computing solutions are not suitable for dynamic, real-time, high-bandwidth, low-latency applications because of issues such as granularity, localisation and configurability; service processing nodes should be distributed and located close to users or to routers or in local data-centres and not only in traditional data-centres.

However, in spite of, or maybe because of, the great achievements that we witnessed in ICT, revenue growth for telecom operators is expected to halve from now to 2020. This means that demand cannot be satisfied by simply increasing network capacity, especially in networks that are becoming always more diverse, dense, mobile and changing unpredictably.

The answer to the challenging and sometimes contradicting necessities summarised above, consists in reducing capital and operating costs, by using low cost technologies, reducing energy consumption, sharing and optimising resources utilisation by dynamically allocating them in time and space, and in general resort to virtualisation techniques as much as possible. Benefits of a full virtualisation of network devices, at all layers, include: i) sharing: resources divided into multiple virtual pieces used by different users; ii) isolation: sharing of a resource does not endanger security and privacy of users; iii) aggregation: if resources are not big enough to accomplish a task, they can be aggregated; iv) dynamics: reallocation of resources in space and time on demand; v) ease of management: software-based devices are easier to manage and update.
In addition, it is necessary that the network be programmable, as a function of the needs of the services that it provides. An example of the capabilities of a virtualised and programmable network is the concept of a network slice; a virtual, end-to-end network, deployed in software, which runs in parallel to other slices on a common hardware infrastructure. A network slice also allows the isolation and support of different classes of services/customers.

The overall vision is thus the one of a software network with an application/service-centric network control able to dynamically share and allocate virtualised resources, allowing to: reduce costs, simplify network management, increase flexibility, ease evolution, and dynamically deploy network services.

5G will be, then, a fully “softwarised” network providing fixed and mobile Ultra-Broadband access to a distributed cloud infrastructure.

The SUPERFLUIDITY project contributes to the vision of a “superfluid” 5G network, which will have the ability to instantiate services on-the-fly, run them anywhere in the network (core, aggregation, edge) and shift them transparently to different locations. Such capabilities are a key part of the converged cloud-based 5G future - they will enable innovative use cases in the mobile edge, empower new business models and allow almost instant roll-out of new services, and reduce investment and operational costs.

To this end, the SUPERFLUIDITY project tackles crucial shortcomings in today’s networks: long provisioning times, with wasteful over-provisioning used to meet variable demand; reliance on rigid and cost-ineffective hardware devices; daunting complexity emerging from three forms of heterogeneity: heterogeneous traffic and sources; heterogeneous services and needs; and heterogeneous access technologies, with multi-vendor network components.

The SUPERFLUIDITY solution is based on: a decomposition of network components and services into elementary and reusable primitives; a native, converged cloud-based architecture; the virtualisation of radio and network processing tasks; platform-independent abstractions, which permit the reuse of network functions across heterogeneous hardware platforms, while catering to the vendors’ need for closed platforms/implementations; and high performance software optimisations along with leveraging of hardware accelerators.

As a result, the 5G network will benefit from: i) location-independence: network services deployable in heterogeneous networks; ii) time-independence: near instantaneous deployment and migration of services; iii) scale-independence: transparent service scalability; and iv) hardware-independence: development and deployment of services with high performance irrespective of the underlying hardware.
More details, and the specific project objectives, can be found in the public deliverable D1.2: Project Vision and Roadmap, v1 (http://superfluidity.eu/results/deliverables/).

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

"SUPERFLUIDITY builds upon the recognition, supported by a break-down analysis of use cases, that network services and functions usually comprise multiple elementary primitives or building blocks, which we generically call Reusable Functional Blocks (RFBs), and which can operate at different layers in the protocol stack. The refactoring of traditionally monolithic network services or functions in the form of a modular composition of ‘more elementary’ RFBs leads to a main key strategic advantage: decoupling of the network service or function logic (programmatically defined via a platform-agnostic configuration ‘language’ or ‘script’, formally specifying block chains and/or abstract extended state machines), from the actual underlying implementation of RFBs (vendor-specific and possibly leveraging different HW platforms and accelerations). Such main key advantage brings about a number of specific assets, including but not limiting to: i) reusability of blocks across different functions or services and hence significant cost savings in the design of such functions, ii) portability and mobility (instant deployment) of services and functions across different platforms and at different points in the network topology, iii) simplified provisioning, at RFB level rather than at the service level, iv) ability to leverage different RFB implementations on different platforms (and with different performance).

The main high-level innovative characteristics of the project so far include:
• The notion of RFB. Our definition of RFBs is on purpose very broad and not bound to a specific ‘size’ or ‘type’. By this approach, we retain generality and permit backward compatibility casting the RFB concept into already existing standards and components/functions (hence further including ‘complex’ RFBs which, even if they could in principle further benefit of a functional decomposition, they are ‘de facto’ marketed as monolithic components). This said, we concretely envision at least two ‘levels’ of RFBs: ‘higher level’ functions, such as those envisioned in the scope of current NFV standardisation and which are chained together to build complex services, and ‘lower level’ primitives which are envisioned within domain-specific network devices (such as OpenFlow-type actions for core network devices, signal processing blocks in SDR/C-RAN, etc.) and relevant standardisation bodies. In order to maximise impact, we are considering different Standards Developing Organizations (SDOs) in this scope, which range from the ETSI-NFV and MEC ISGs to the SFC working group at IETF and the NFV research group at IRTF. In our working assumptions, RFBs can be expressed in terms of other RFBs, in theory with no limits to the recursion levels. Suitable languages to support this approach needs to be defined; we are considering the use of the NEMO network modelling language to support RFB description and combination and an internet draft has been submitted.
• Controlling and placing RFBs. The identification of a suitable set of RFBs, which can act as basic building blocks is not sufficient: indeed their coordination, orchestration, and (co)operation to provide a value-added service require supplementary techniques, systems, and computational environments. Specifically, innovation and even foundational research is needed in terms of i) languages, formalisms and mechanisms to specify a desired (composed) service, and validate a given configuration (this entails the ability to semantically describe each RFB so as to guarantee functional equivalence among different implementations), ii) algorithms and data models for network resource description and functions’ placement, and iii) environments and tools to run-time enforce, provision, and reconfigure, a desired composition.
• VNF orchestration advances. Considering the ETSI-NFV architecture, the RFB concept maps into VNF and VNF components (VNFC). The work on VNF orchestration is already somewhat consolidated (or at least launched) and SUPERFLUIDITY is contributing in this arena and in tight relation with emerging systems and standards. The innovative contributions consist in the automatic characterization of VNF performances based on measurements, in the evaluation and monitoring of service KPIs, in the automatic scaling mechanisms for the dynamic allocation of resources and placement of VNFs. The integration of MEC (Mobile Edge Cloud) into the same orchestration framework, including data plane aspects (traffic offloading function) is also innovative. The project has also addressed the decomposition of C-RAN into VNFs and its implementation with a Docker-based solution cooperating with an SDN controller.
• Packet-level RFBs definition and composition. The ability to describe and run-time enforce stateful compositions of elementary building blocks at the level of high performance network nodes, and while challenging line rate operation, is still in its infancy and requires disruptive and novel research efforts, which SUPERFLUIDITY is addressing. For instance, ONF OpenFlow router-level data plane configurations, i.e., flow tables, are as of now static, and the need to rely on slow-path CPU for stateful processing or dynamic/adaptive flow table reconfiguration, or even decentralize the stateful logic in remote controllers may severely impair performance and applicability to latency-critical scenarios and load the network with extra device-to-controller signalling burden.
• Virtualisation on diversified HW platforms. Virtualisation technologies play a key role in SUPERFLUIDITY. We specifically envision a number of complementary major research directions in this area. On one side, SUPERFLUIDITY is working to significantly improve virtualization performance and efficiency on commodity computing platforms (X86, ARM, etc.), also using emerging approaches such as unikernels or containers, so as to significantly raise the performance bar in terms of scaling, throughput, latency, VM consolidation, and so on. We are designing and implementing benchmarking tools to analyse the performance of software and hardware accelerations for packet switching over commodity hardware. Additionally we are also investigating the resource request rationalisation in order to meet service SLA’s and associated KPIs. The approach being taken focuses on parameterisation optimisation for a given service deployment, in a heterogeneous NFVI, in an automated manner, to support the necessary levels of scalability. On the other side, SUPERFLUIDITY is exploring virtualisation over domain-specific network processing and signal processing HW architectures relying on commodity (but domain-specific) HW, so as to gain the performance advantages of domain-specific hardware acceleration, while retaining re-purposability and programmability of general purpose platforms. In particular, we are considering the design and implementation of C-RAN baseband stack employing a dynamic dataflow paradigm. Finally, a further challenge lies in identifying the programming models and resource descriptions that permit seamless adaptation of a service description to the underlying hardware, so as to match a desired performance and scaling target.

In summary, SUPERFLUIDITY’s innovative assets are mainly revolving around the following aspects:
• Management of heterogeneity and hardware diversity, not just hiding it, but taking advantage of it
• Considering all network segments/domains, from the core to the edge, including the radio segment, not just one/some of them (this is also related to the heterogeneity issue)
• Emphasis on programmatic construction (building) of complex services out of simpler building blocks
• Emphasis on reuse
• Function/Service portability, not just flexible placement
• Orchestration, as this is a key enabling technology required to enable automated deployment and management of networks services at scale in heterogeneous and distributed NFVI environments.

The work performed during the first year of the project can be very briefly summarized as follows, for each Work Package (WP):
• WP1 (Project Management): establishment of the Project Office and of procedures for monitoring the workflow of the project and of procedures for management; establishment of administrative and reporting procedures; definition of standard formats and forms for project documentation. Implementation and management of all required technological infrastructure for supporting the project (web site, repositories, mailing lists, conference calls tools); organization of meetings and conference calls. Strategic and Technical coordination; anticipation of integration activities, establishment of an additional project testbed and management of the related funding via a budget shift among partners. Management of the cooperation with the 5GPPP and related working groups and projects.
• WP2 (Use cases, System Requirements and Functional Analysis): analysis of potential use cases for a superfluid network, business and technical requirements, consideration of their likely impact on the SUPERFLUIDITY architecture, and analysis to decompose network functionality into reusable components.
• WP3 (Cloud-Native Edge System Architecture): definition of the SUPERFLUIDITY architecture, based on the concept of Reusable Functional Blocks (RFBs), which is applied to different heterogeneous RFB Execution Environments (REE). Standardization convergence: harmonization of MEC (Mobile Edge Cloud) and C-RAN over an Extended-NFVI that represents an evolution of the ETSI NFVI concept.
• WP4 (Heterogeneous Infrastructures and Abstractions): i) development of an automated KPI mapping methodology for services, with a framework for automated deployment of the service under test, execution of test cases, automated collection, processing and modelling of the data; ii) study of modelling approaches for measuring and gathering KPI; iii) implementation of a flexible and modularised C-RAN baseband stack, employing a dynamic dataflow paradigm; iv) benchmarking of software and hardware accelerations for packet switching over commodity hardware.
• WP5 (Virtualisation Platform Implementation and Network Dynamics): i) implementation and evaluation of function allocation algorithms, with the development of a low level component allocation framework over the FastClick modular software router; ii) survey and measurement of available virtualisation technologies implementing a number of radical optimizations to existing virtualization platforms (e.g., Xen and KVM) in order to have VMs exhibiting properties more closely associated with containers but without suffering from their isolation and security issues; iii) modelling, analysis and tuning of VIM (Virtual Infrastructure Managers) performance.
• WP6 (System Orchestration and Management Tools): i) identification of requirements and gaps for the control framework; ii) proposal for the use of the NEMO network modelling language to support RFB description and combination and to express SLA requirements; iii) analysis of resource allocation and placement problems in different contexts; iv) modelling and design for Symbolic Execution tools.
• WP7 (System Integration and Validation): the System Integration and Validation WP start has been anticipated to Jun 1st 2016, coordinating current development and integration work; in that context, two testbeds have been established, ahead of schedule: i) testbed #1 is located at Nokia France premises. It is a hardware and wireless platform allowing to demonstrate some innovations (e.g. Cloud RAN, RFB decomposition,...) conducted in the project and made accessible to all partners; i) testbed #2 is a hardware platform located a BT UK, consisting of 5 servers and a switch, and made accessible to all partners, which allows flexible virtualisation experiments and demonstrations to be run by all partners. These will allow various aspects developed in the other WPs to be explored earlier. As of now the project decided to prepare and show 4 different demonstrations, titled: Orchestration, Software Defined Superfluid Wireless Network, Verification of OpenStack, Mobile Edge Computing. Both testbeds and demonstration are described in the public deliverable D1.2: Project Vision and Roadmap, v1 (http://superfluidity.eu/results/deliverables/).
• WP8 (Communication, Dissemination, Standardisation and Exploitation): Communication, dissemination, standardisation and exploitation activities, reported in the public deliverable D8.2: First report on Communication, Dissemination Actions, Standardization and Open Source Contributions (http://superfluidity.eu/results/deliverables/).
More details on the work performed in each WP can be found in the public deliverables available at http://superfluidity.eu/results/deliverables/.

The main achievements of the project so far are listed below:
1. Definition of the SUPERFLUIDITY architecture, based on the on the concept of Reusable Functional Blocks (RFBs)
2. Development of an automated KPI mapping methodology for services into VNF performances
3. Modelling and design of Symbolic Execution tools for packet processing functions; semantic RFB description using the SEFL language
4. Proposal for the use of the NEMO network modelling language to support RFB description and combination and to express SLA requirements
5. Implementation and evaluation of function allocation algorithms for packet processing functions in a modular software routers architecture
6. VNF and infrastructure telemetry framework that works in conjunction with a workflow engine to facilitate instantaneous life cycle management operations, such as migration or scaling.
7. Design and implementation of a MEC (Mobile Edge Cloud) architecture integrated in the SUPERFLUIDITY vision. Modular MEC prototype using SDN/NFV technologies
8. First version of a decomposed C-RAN solution prototyped, including front-haul and interacting with a decomposed core network. The networking of the different RFBs, deployed as a Docker, is managed by an SDN controller.
9. A flexible and modularised C-RAN baseband stack has been implemented employing a dynamic dataflow paradigm
10. Benchmarking of a software switching, which was leveraged for improving resource allocation (e.g., queues and cores) to the software switching, as well as for modelling the cost of network switching for different SFC deployments.

In the figure, we show the SUPERFLUIDITY architectural framework and the above main achievements drawn over it, to give a general idea of how they are related to the main architectural components.
In Figure 1a, the top layer includes the different components involved (CRAN, MEC, virtual core and Data Centres (DC)), while in the bottom layer the different types of physical DCs are shown (namely Cell-site, Local, Regional and Central). The right side of the diagram reports the management and orchestration components. The SUPERFLUIDITY architecture is drawn in between such axes. For a detailed explanation of this figure please see section 1.2.3 below in this document.
In Figure 1b, we report the ten achievements over the architecture diagram of Figure 1a, with no claim of rigorousness; in facts, most of the achievements actually refer to the management and orchestration components in the right side of diagram and should be also represented there, making the diagram unreadable."

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)

In the following, we outline the project’s progress beyond the state of the art in a set of research areas that we believe are more of importance to SUPERFLUIDITY. This progress is the one expected from the whole duration of the project. More detailed and current information about the project’s progress beyond the state of the art in the first year of work, as well as the project’s innovation potential, can be found in the public deliverable D8.3: Innovation and Exploitation Plan (http://superfluidity.eu/results/deliverables/), which reports the advancements reached as of today.

• Cloud Networking: SUPERFLUIDITY will aim to meet the stringent requirements imposed by future 5G networks by designing and implementing a superfluid, converged network architecture that is location, hardware and time-independent. The work will push the boundaries of what is currently possible with virtualised, software-based packet processing (10-40Gb/s and higher, extremely fast service instantiation and migration in milliseconds, massive numbers of concurrent virtualised services on a single platform, significant power reductions, etc.). The goal is to bring the advantages of cloud and software-based networking to 5G networks so that services can be deployed whenever and wherever they are needed, and to leverage the availability of inexpensive, off-the-shelf hardware in the process.
• Network Services Decomposition and Programmability: SUPERFLUIDITY will devise programming abstractions specifically targeted to 5G functions. The API design work will address three programming levels: service, function, and processing levels, and will attempt to maximise viability by reusing existing standard (when applicable) or research community best practices. Work will on one side target the definition of 5G specific actions and events, and on the other side will address the specification of the constructs needed to combine and orchestrate a desired execution of such actions (conditioned on the arrival of events). Particularly promising and forward-looking is SUPERFLUIDITY’s approach of combining block-based composition abstractions (such as those exploited in Click routers, in some software defined radio architectures, or emerging in the ETSI NFV work on service chaining) with event-driven programming paradigms such as basic match/action based approaches or more powerful stateful abstractions based on extended finite state machines.
• RAN Cloud and Mobile Edge Computing: Beyond the current vision of a static RAN function fully located in one “edge computing” place, SUPERFLUIDITY will support the ability to modularly “hot” replace eNB functions (such as scheduling) and to permit migration of such functions between edge clouds and the antenna subsystem, so as to balance algorithmic complexity with front-haul capacity. SUPERFLUIDITY will also transcend current Mobile Edge Computing vision where non-RAN functions (local caching, CDN, etc.) are envisaged to be co-located only at the eNB by enabling their migration between the RRH and the edge cloud, to maximise their performance.
• Automated Security and Correctness: SUPERFLUIDITY will provide a two-pronged, complementary approach to security. First, it will go beyond the state of the art, providing a pre-deployment checking system that will ensure that virtualised network services do not negatively affect the network nor other tenants; unlike approaches in the literature, the system will be both scalable and stateful, able to model most types of services. Second, SUPERFLUIDITY will implement a post-deployment system that will learn the behaviour of traffic and detect any anomalies, thus providing a further security mechanism in cases where the checking system does not have information about the processing performed by a network function, or when static analysis is inaccurate.

As regards the expected potential impact of SUPERFLUIDITY, the project is at the end of its first year; therefore, most of the results obtained so far will be exploited internally in the remaining project activities. Anyway, we have selected few results that are more mature or show a high potential to be exploited outside the project activities; these are listed below:

• Development of an automated KPI mapping methodology for services into VNF performances
• Extensions of the NEMO network modelling language to support RFB description and combination and to express SLA requirements
• VNF and infrastructure telemetry framework that works in conjunction with a workflow engine to facilitate instantaneous life cycle management operations, such as migration or scaling
• Modular MEC prototype using SDN/NFV technologies
• Prototype of C-RAN front-haul and core network based on Docker containers.

Related information

Record Number: 192911 / Last updated on: 2016-12-15