I wish starting this post by resuming the vision of Telecommunications as a gigantic supercomputer nicely described in Roberto’s post. Internet and Telecommunications today are based on massive distributed networks interconnecting processing and storage nodes. This is recalling the image of hundreds of thousands of chips interconnected each other in a supercomputer: even more amazingly, in the near future, both inter- and intra-chip communications will be carried through optical signals. This is also what we’re dreaming for future transport networks as “photons are faster and consume less than electrons”.

Let’s make an example, looking at a data center, which can be seen as a sort of a supercomputer. The data center network fabric is a network capable of interconnecting thousands of server, storage and other network ports in a flat, ultra-low latency, high bandwidth infrastructure that provides any-to-any connectivity. A flat fabric-based network architecture eliminates the need for multiple layers, switch-to-switch interactions: it simplifies network management and operations while improving performance. Other nodes can be seamlessly added given the fabric’s high degree of scalability. Also, we may say that data center approach  is application driven, rather than network driven. Indeed this is what we’ll see in the future also for Telecommunications networks.

In the future, end Users (applications) will be more and more able to “drive the network dynamics”, introducing flooding the network by the edges: by Users it is meant not only people by also machines, smart objects, things and any device which is attached to the network at the edge. In fact, technology advances (e.g. standard h/w performance, embedded communications, device miniaturization, etc.) and the related costs reductions are progressively moving an incredible amount of processing, storage, communications-networking capabilities at the edge of traditional networks, i.e., towards the hands of the end Users. Actually, it’s a few years that we are witnessing this trends: examples are CDN delivering contents from caches at the edge, closer and closer to end Users, other edge computing services (e.g., web acceleration) provided by Players like Akamai, etc.

New paradigms as SDN (Software Defined Networks) and NFV (Network Functions Virtualization) are creating the conditions to reinvent networks architectures as they are offering the possibility to look at the whole network in an abstract way, shaping resources and connecting them in a dynamical way.

In this post, for example, I’ve already pointed out that chance of virtualizing network and service functions which are provided today by expensive L4-L7 functions middle-boxes, and moving them in the Data Centers or even better at the edge (where this huge amount of resources is accumulating), as closer as possible to the Users. This will be a big change. Middle-boxes are today closed pieces of equipments. Not only said stateful middle-boxes are breaking the end-to-end principle, they also contributing to the network ossification, but they are representing a significant fraction of network capital and operational expenses (due to management complexities).

My bet is that in the near future, the edges will look like a data center network fabrics capable of interconnecting thousands of standard hardware servers, storage and other network nodes. Edges will become like Distributed Network Computing Platforms (creating the so-called Edge ICT Fabrics), composed by pools of general purpose h/w resources (capable of computing, storage and network I/O). Edge ICT Fabric will includes Users’ devices, CPE, aggregation nodes, Edge PoPs (which can be even seen as micro-data centers at the edge). Edge ICT Fabrics will be characterized by high flexibility, performance and self- adaptation at run-time (e.g. dynamic flocking of resources according to needs). Importantly, it will be also possible harnessing and combining all unused resources (e.g. computing and storage power at end Users’ home and in the edge micro data centers). Through the Edge ICT Fabrics, it will be possible programming, allocating and moving a variety of virtual architectures (spanning across diverse edge networks or even across today Data Centers) on-demand, based on Users’ applications, also meeting governance and biz requirements (no more ossified networks structures).

This is a change of paradigm: a storm of pieces of software (even open source) executed on general purpose hardware will allow to abstract all network functions and services, in a way to impact profoundly Telecommunications and ICT business. This impact should be considered from the point of view of Incumbents’ networks, but as well as from the one of OTT, enterprise networks and consumer electronics.

Fog Computing vision goes beyond Cloud Computing by arguing the use a sheer number of resources distributed at the edge of the network. Apparently another buzzword. On the other hand, today we are already witnessing a progressive migration of processing power, storage capability and embedded communications towards the edge of the network; this trend, coupled with devices miniaturization and costs reductions, will create the conditions whereby Users will literally “decide and drive”  future ICT networks and services. This will have big impacts. This floating “fog” of ICT resources at the edge will give rise to new biz models based on new forms of competition and cooperation between existing Providers, and new  ones entering the arena, including utilities, car manufacturers, consumers’ electronics, public administrations, communities, etc.

In these dynamical games we’ll see innovative proposals, rewarded directly by the market itself, which will be essential encouragement for further investments. So, ideally, in the edge ICT fabric it will be possible creating, programming, instantiating or migrating dynamically different types of virtual functionalities and services as well as alternatives of the same. No more ossified architectures. In other words sort of ephemeral networks of resources will plastically self-adapt to humans’ dynamics. And it shouldn’t be a surprise discovering that this follows the laws of emergence in “complex systems”!

Emergence of flocks of birds: each individual responds to local conditions with a similar rules set.

Emergence is topic already acquiring a growing interest in social networks: there are interests in modeling and predicting the dynamics of groups of people, the viral diffusions of certain ideas or concepts, use of resources, or even the potential adoption of product and services.  Think about the the convergence of Internet and the social attitude of humans: beyond sites such as Facebook, LinkedIn, MySpace, Wikipedia, YouTube there is a broader process to form connections with others, build groups and to engage communications.  A political message, or a piece of news or a meme are examples of information that can spread from person to person, in an epidemic way. This can catch the attention of millions of people creating ephemeral human dynamics, made visible by on-line expressions.

 Modeling ICT social epidemics provides the opportunity to identify influential behaviors, or ways to predict, trigger or incentive mass adoptions of products or ICT services. Several mathematical approaches have been proposed for modeling these dynamics: a diffused idea is modeling the state of each person as a member of a lattice and updating it by using simple rules depending on the states of neighboring members. Each state could be represented by a set of variables such as cultural skill, preference, beliefs, etc. and each of them associated with a certain flipping energy, which is the cost of changing the state given its connection with other members, which are in other variables-states. At the end of the day, it’s about how environment conditions, or messages, will influence people and vice-versa how individuals influence each other and the environment.  In a next post I’ll provide some simulations examples.

 In summary, as in any complex system, also in the Edge ICT Fabric, it will a matter of taming complexity and extracting simplicity out of “local-to-global” dynamics.

“Fire together, wire together”  is a famous paraphrase of Donald Hebb’s postulate summarizing the current understanding of how learning and memory could be realized on a cellular level. Formation of new memories is accompanied by an increase in the number of synaptic structures: two neighboring neurons showing correlated activity should form or strengthen a synaptic connection between them. In other words: if two neurons on either side of a synapse are activated simultaneously (i.e. synchronously) then the strength of that synapse is selectively increased; if two neurons on either side of a synapse are activated asynchronously, then that synapse is selectively weakened or eliminated. This ability of neurons to modulate the strength of their synaptic connections depend on the relative timing of pre- and postsynaptic action potentials: this synaptic plasticity, called spike-timing-dependent plasticity (STDP), has become an attractive model for learning.

Schematic representation of Hebb’s postulate

Let’s try applying this interpretation of Hebb’s postulate to a network, composed by a sheer number of nodes (like neurons, in the metaphor), e.g., future edge networks (as described in former posts). Each device reflects the behaviors of a User, which is somewhat stochastic and unpredictable, at least when it is viewed at the microscopic level of individual acts (as the neurons).

We may wonder about the emergent behavior of an edge network environment can be viewed in a similar way, as a dynamic game of neurons, influenced by the degree of coupling between the actions of different Users, (or Players at the business level). In fact, in principle, these couplings may have the effect of amplifying (logical) local interactions in a manner analogous with Hebb’s postulate of learning.

Like in the neurons networks of the brain, self-organization takes place in the form of a feedback loop: on one side certain activity patters are produced (by a given neuron network) in response to input signals, on the other side connection strengths of the network are modified in response to the activity patters (e.g., in the brain due to synaptic plasticity). When the feedback between changes in the connection strengths and the changes in the activity patterns are positive we get self-organization (instead of stabilization).

We may conjecture that the emergent behavior of a complex network of Users could be explained in terms of stochastic actions of each User, as seen individually by the other Users. This is in line with what Ettore Majorana wrote in his paper “The value of  Statistical Laws in Physics and in Social Sciences”, observing that by experimental set-ups one can prepare a complex chain of phenomena started by the accidental disintegration of a single radioactive atom: …nothing forbids us considering as plausible that human events too can be originated by an equally simple, invisible and unpredictable occurrence. But, in spite of this unpredictability, self-organized learning is still achievable, as shown by our brain !

Imagine an edge network capable of self-organized learning, i.e., capable of learning from the environment and, through learning, capable of improving the performance. This without a Teacher, i.e., an Operator managing the network. This means embedding, into its nodes and devices, specific algorithms, with local rules, able to learn to compute input-output data mapping with desired actions or properties (local rules means rules capable of changing for example the “strength” of the interconnections (synaptic weight) of a node (neuron) in the immediate (logical) neighborhood. This is fire together, wire together.

Let me resume Roberto’s last post. In this piece of news, Intel is claiming working on 48-core chip for smartphones and tablets. In five-ten years this technology – they say –  “could really open up our concept of what is a smartphone, a computer…”. Well, already today, computers and even some small mobile devices use multi-core chips, but 48 or even hundreds (in future 3D chips) in another story. Surely, one issue is making sure that software applications can take advantage of so many cores, and before that, even more the need of modifying how operating systems are developed, making them far more parallel. Nevertheless they say they sure that when hardware is ready, the software will be, too.

These advances could really open up also our concept of edge networks. It is likely that, in the near future, technology progresses (coupled with the down-spiraling of costs) will make available an incredible abundance of processing and storage capabilities at the edge of current networks (where actually “intelligence” has migrated) as provided by local resources and Users’ devices. These trends will transform edge networks into a general purpose network infrastructure with multiple interacting domains (operated even by diverse Players), where slices of virtual networks will be assigned dynamically to different entities and Users, according to needs. Most of network functions will implemented in software running on top of low cost general-purpose hardware.

Abstract view of a software router constructed as a cluster of servers (Routebricks)

In this paper (and in former one), for example, it is claimed the feasibility of high performance software routers, thanks to the recent advances in server technology that enable high-speed parallel processing. Interestingly they demonstrated a 40Gbps software router architecture that parallelizes router functionality both across multiple servers and across multiple cores within a single server. Router capacity can be linearly scaled through the use of additional servers and it is fully programmable using Click/Linux environment. Servers hardware is low cost general-purpose.

Still in this direction, this nice paper introduces the ClickOS, a minimalistic network operating system that runs on top of a Xen hypervisor and that is based on the Click modular router software.

Advances in embedded communications and computing technology, development of routing, packet processing (and other net functionality) in Open Source software, running on general-purpose hardware, will dramatically transform edge networks into general purpose distributed data centers with plenty of local resources (operated even by diverse Players).

This means that Networks and Clouds will become commodities and - as Roberto said in his last post - that we are moving towards a loss of perception of telecommunications infrastructures. In turn the ambient will acquire a sort of nervous systems made of a sheer number of termination points offered to the Users. Imagine the opportunities behind this amazing complexity!

Computing is becoming low cost and pervasive, embedded in any node, Users’ device, everyday object and in the physical environment (sensors, actuators, etc) as well. An open and dynamic networked world will be the arena of next services and applications. Traditional control and management approaches will be ill-suited to face such environments: the question is how effectively exploiting coordination in huge ensembles of distributed autonomous entities (against strict requirements such as dynamism and complexity). Forget also traditional middleware: communication and computation costs would be too high, and solutions brittle and fragile. Feasible approaches to control and manage myriad of interacting nodes and devices are still unknown…but do we really need them? Let’s change the perspective: did Nature invent a way to control the behavior of any single neuron ? Not indeed! Part 1 of this post (drawing inspiration from neurotransmitter functioning in neuron networks) has proposed a vision of future networks where each node, device, machine, smart object (like a neuron) is capable of autonomous local self-adaptation reactions to the context (through neurons’ interconnections) and where a global harmonization is made through viral propagation of coordination and context information (as neurotransmitters do). This Part 2 is proposing a simple proof-of-concept aiming at demonstrating that this is feasible even today.

Imagine Users (cars, kiosks, lamp streets…) having a sort of communication halo around them: Edge Networks (see a previous post) will emerge spontaneously (as flocks of birds flying around) through these halos overlapping cross-interactions. Imagine each node having perception of its local context, the environment inside its halo. Each node diffuses its context information hop-by-hop accordingly to certain propagation rules. Any context can be accessed locally but at the same time it takes account of the influences of the context and coordination information propagated from other nodes (also the fixed ones). We’ll have a sort of global coordination field, injected by nodes in the network and autonomously propagating … like neutransmitters. In other words, nodes are interacting with each other and with the environment by simply generating, receiving and propagating distributed data structures abstracting context information. This field is providing nodes with a global representation of the situation of the overall network, which is immediately usable, like an object moving in a “gravitational” field. Environmental dynamics and nodes local decisions will determine changes in the field closing a feedback cycle. This is enabling a distributed the overall self-organization.

In real proof-of-concept nodes’ halos can be easily implemented with a smart phone (acting as Wi-Fi Hot Spot), one (or more) cheap, tiny PC (e.g. a Raspberry Pi for $ 25) and one (or more) microcontroller (e.g. based on Arduino). Coordination field can be made of “tuples” of data which can be injected and diffused by each node. Local reading of these “tuples of data” (e.g. through pattern matching) can trigger local self-adaptation behaviors. Plenty of open source applications are available on the web to implement nodes primitives and local autonomic behaviors. It’s simpler than expected.

Future Networks: local reactions and global self-organization

Surprisingly, if we look at the network dynamics as a “many body problem” we can even define the Hamiltonian (a sort of energy function describing the state) of the network. Just following Nature.

I’ll elaborate that in a next post.

Users’ devices (e.g. smart phones, with ever growing storage and processing capabilities, acting as hot-spots), and a multitudes of smart objects and things (e.g. from Consumers’ Electronics) with embedded communications, will create new challenges and opportunities at the edge of the network. It is estimated that, in less than ten years, there will be a few hundreds of billions of electronic devices (including machines, sensors, actuators, etc) connected with each other and to the Internet. A wave, innovating networks, starting at the edge.

At the edge (in the last few meters) there will (soon) be a growing number of such communicating entities, with powerful storage and processing capabilities, interacting one each other locally. Imagine cars or Users having a sort of communication halo around them (i.e. a range of connection, interaction). Imagine also kiosks and lamp streets having their own halos. Overlapping halos will allow networks to emerge spontaneously (as flocks of birds flying around). Short-middle range connectivity will be covered by local device-to-device communications, whilst long range interactions will be enabled by hopping into the big Net. Services and data will be virally delivered through multiple devices, machines, objects mostly by using local resources.

Is this scenario so far away in the future ? Not really: some military solutions (almost ready for civilian needs) are already available.

Today, creating – your own “halo” – by yourself would cost you less than one or two hundred euros (obviously depending on what and how many devices you wish to have). You may like to include, for example, an Android smart phone (which can act also as Wi-Fi Hot Spot), one (or more) cheap, tiny PC (e.g. a Raspberry Pi for $ 25) and one (or more) microcontroller (e.g. based on Arduino) for controlling any sensor, actuator or electronic gadget.

Raspberry Pi: a cheap, tiny PC for $ 25

It will be like a fully fledged wireless personal area network (with thousands of free applications available on the web). Once equipped with autonomic features it will be ready to interact spontaneously with other people’s halos to create dynamic local networks. Welcome to Edge Networks.

This scenario raises important issues for Stakeholders to consider. Are we ready?

Future networks will be different from what an engineer would think today. I’m an engineer and I’ve experienced ways to design, to optimize and to manage network resources in order to guarantee QoS. But imagine plenty of tiny smart nodes, devices and objects interconnected by network of networks thus weaving themselves into the fabric of everyday life to provide any sort of services. Imagine plenty of viral networks at the edge where collective behaviors emerge from the interaction of large numbers of such nodes adopting very simple local rules.

Viruses behaviors to inspire Viral Networking

Edge networks will become robust by definition, but potentially subject to state transitions and the traditional end-to-end QoS issues will assume other perspectives: the network arena will be transformed into dynamic games of many (also new) Players, not only Telcos. Networks dynamics will be governed by the mathematics of “Chaos”: different cooperation-competition strategies can be take place as interactions of states attractors, or even better, basins of attractions in a network phase space. By the way, this an image that can be applied also to the neuron networks in a brain: huge amout of entities, embedding communication, storage and processing capabilities, interacting with each other with simple rules. Then, it is like in Nature, at least metaphorically: no central control, but evolution, is managing complex networks. But evolution normally takes a long time. On the other hand, we have technologies which allows us mimiking a dramatic acceleration of the time variable, and markets will make the natural selection of the best viral solutions.

At the end, it will be possible by tuning very simple autonomic rules of plenty of nodes to guide the network dinamics according to predefined goals and strategies. The degree of “autonomicity” of such viral nodes will provide the basis for the concept of controlled self-organization.

This set of simple rules (mimiking their nervous system) of tiny smart nodes, devices and objects represents a sort of “link” between (generalised) sensors and actuators. Each node, through sensors, perceives its environment, detects the existence of other goals and, through actuators, put in place the required actions. Another similar way to see it is a reflexive coevolution of behavior and structure, which is typical of what they call adaptive networks (e.g. the Web).