06 October 2016

Towards Artificially Intelligent Networks

Yesterday I've made my keynote at the IEEE Conference CloudNet2017, by the way very interesting event with high quality publications and presentations. 

Here it is the abstract of my talk:


The increasing levels of flexibility and programmability provided by the exploitation of enabling technologies such as SDN-NFV-Edge/Fog, leading to the Network Softwarization, will determine in turn higher levels of management and control “complexity”: in fact, rather than operating sets of closed physical nodes and systems, it will be necessary allocating and orchestrating a dramatically higher number of software processes, logically intertwined and dynamically moving in the Telecom infrastructures. Central Offices will become like Data Centers.

In the X-as-a-Service era the business sustainability passes through ways for increasing QoE, reducing Time To Market and guaranteeing cyber-security

This “complexity”, outstripping human control and operations ability, could be tamed only by exploiting, in the real-time Operating Platforms, leveraging on Applied mathematics, Artificial Intelligence methods and systems capable of making “actionable” the infrastructure Big Data (e.g., logs, alarms, and other data). 

In fact, it is easy to predict that key question of the ongoing Digital Business Transformation is automating the Operations processes (today's OSS-BSS), from the management to the control to the orchestration of physical and logical resources 

...but this is paving the way towards Artificially Intelligent Networks, where SDN-NFV-Cloud-Edge-Fog Computing will converge with A.I. systems and methods.

Happy that the keynote has been well received with several questions!
Who is interested in the presentation, please drop me an email, please to share it.

Very proud to show the appreciation.
Thanks to the General Chair (Prof. Stefano Giordano) and to the Organizing Commettee !

04 October 2016

Quest for Real-time Operating Platforms

Digital Business Transformation mean a number of objectives for Network Operators and Service Providers: saving Operational costs, reducing Time to Market, improving de-commissioning procedures, improving the “quality of service” but also becoming ready for providing new ICT services, even those which are still unpredictable today. In a sentence, Telecommunications infrastructures should become “good enough” to be economically sustainable in highly dynamic and changing scenarios.

We know that SDN-NFV are considered today two of the most promising enabling technologies to achieve these goals. However, the target of increasing the levels of flexibility and programmability will determine, in turn (reverse side of the coin), higher and higher levels of management “complexity”: in fact, rather than managing sets of closed physical nodes and systems, it will be necessary allocating and orchestrating a huge number of software tasks, logically intertwined and dynamically moving.

There is also an overall consensus that 5G production (and beyond) environment will look like distributed clouds of IT systems, interconnected through ultra-low latency (radio and wired) connections, capable of executing software processes and applications, dynamically meeting Customers’ needs. As a matter of fact, already today we’re witnessing the interweaving of technologies such as Edge and Fog Computing with SDN and NFV.  And “Softwarization” will allow decomposing the network and service functions into chains of software tasks. End-to-end service provisioning will require that this functional decomposition will be followed by an optimal allocation and orchestration of the virtualized functionalities across User Equipment, RAN, Mobile Edge and Core resources. Eventually, this will bring a unified service modeling whereby SDN services (e.g., controllers), NFV services (e.g., Virtual Network  Functions), and Cloud services are seen as “application” executed on virtualized resources.

TOSCA (Topology and Orchestration Specification for Cloud Applications) will be a natural candidate for the Northbound interfaces of the Real-time Operating Platforms. TOSCA  is a standard from OASIS that targets interoperable deployment and lifecycle management of cloud services. In fact, TOSCA uses the concept of service templates to describe cloud workloads as a topology template. The topology template describes the structure of a service as a set of node templates and relationship templates modeling the relations as a directed graph. Node templates and relationship templates (linking different nodes) in fact specify properties and operations (via interfaces) to manipulate the service components. Moreover, it is likely that the YANG declarative data modeling language will be used both to describe deployable instances of a service (e.g., a VNF) and to configure a network device/element at run time.

Eventually, TOSCA and NetConf /YANG could be considered as complementary instruments: deployment templates may trigger the  NetConf /YANG configurations during the instantiation of a service, whist in the Operations the Real Time Operating Platforms can take over configurations at run time. On the Southbound interface a number of well-known configuration protocols and programming language are getting momentum: OpenFlow, NetConf, P4, etc.

At the same time we’re witnessing a growing diffusion of Internet of Things and Machine to Machine communications are creating also a new generation of non-human Customers’, such as Robots, Avatars and any sort of Artificial Intelligence applications.  This “complexity”, outstripping human control and operations ability, will be tamed only by exploiting real-time Operating Platforms, based on Artificial Intelligence (A.I.) methods and systems, integrating management, control and orchestration functions. It will be necessary collecting, filtering and elaborating the infrastructure Big Data, thus “closing the loop” and making the them truly “actionable” for Operations and provisioning of services.

Real-time Operating Platforms should provide an abstraction layer for switching/networking (e.g., Switch, Ports, Links) and compute, storage resources (e.g., CPU, RAM, Disk, Ports, etc.). This allows applications and developers to request connectivity, storages and arbitrary units of compute power without one having to worry about how this translates to bare-metal, Containers or Virtual Machines.  Eventually this evolution will impact deeply the current value chain: in fact, Telecommunications infrastructures, governed by real-time Operating Platforms, will become a single converged industrial structures covering voice, Internet access and other services a la OTT.

In this big leap towards Artificially Intelligent Networks, a new Community will have to be developed capable of integrating Experts in Computer Science, Telecommunications-ICT, A.I. and Applied Mathematics.

Dynamical Systems Internet based on Feynman Machines

I wish celebrating my 100th post on this blog with a short piece about this amazing paper:

Feynman Machine: The Universal Dynamical Systems Computer

Paper proposes a simple but very innovative model which draws on recent findings in Neuroscience and the Applied Mathematics of  Dynamical Systems. One of my dreams, using the advances on Neuroscience and  Applied Mathematics to re-define the way we see the networks.

as mentioned by the paper, the Feynman Machine is a Universal Computer for Dynamical Systems, analogous to the Turing Machine for symbolic computing, but with several important differences capable of bringing to radically new architectures for machine intelligence.

Unlike the Turing Machine (or any digital computer), the Feynman Machine is not “programmed” in the traditional sense we are used to. The structure of the network, the choice and configuration of regions, and connections to its external world together dictate the functionality and capability of the machine, and the actual performance is achieved by online learning of the structure in the world. Like for the nervous system of living being.

Now imagine extending this paradigm by interconnecting each other Feynman Machines (not digital computer as we have today in Internet) and you'll get a sort of Dynamical Systems Internet based on Feynman Machines. A radically new ways of looking at the Network.

As a matter of fact, the combination of SDN and NFV are creating new "complexity" dimensions that, in principle, could allow this leap, thus potentially opening new frontier for Artificially Intelligent Networks.

This will be part of my talk at IEEE CloudNet on 5th October in Pisa.