Many people have no doubt that s/w and h/w technologies will progress so much to offer (sooner or later) Carriers’ class network solutions à la SDN (Software Defined Network) and NfV (Network functions Virtualization). I tend to agree: it is not about whether or not, but when and how. And when will be earlier than we expect now. The question, which is not that debated yet, is how, i.e., the “strategic side” of the potential adoption of these technologies progresses. Can we monetize all of this ?

It is clear that any promising technology is likely to adopted not only if it is reducing costs (e.g., Capex and Opex savings), and if it is trusted, but also if it is really able to create new businesses. Today, SDN and NfV seems to be “panacea”, solving any problems that Network and Service Providers may have in optimizing their infrastructures. But where is the real value in terms of new business? Many say “programmability”: I’m not fully convinced. It’s more, “adaptability”, to me: bringing the network on the top. I think that the point is changing perspective in taking the strategic biz advantages of the tremendous technology advances which we are witnessing: it is likely that in less than five years it will be possible to develop L4 to L7 network functions (almost) fully in software (e.g., executing them in Virtual Machines) and dynamically allocating and moving VMs over distributed resources. Throughput will be improved (as simply it will be possible moving seamlessly virtual resources closer to them, thus reducing RTT) whilst Operators will be able managing QoS policies at higher levels (than we are doing today).

This is like saying that the “mind” of an “ossified network” could be removed from today’s closed (and costly) boxes and moved up to a software flexible level, where is orchestrated, thus increasing dramatically network “adaptability”. And the network will be then “on the top” ready for new biz models, or redefining the current ones.

Please note that adaptability is a broader concept than adaptation: it’s the capacity to continue to function in an unknown or uncertain environment (as from Michael Conrad’s definition), by altering its structure and dynamics. The constraints that keep a network capable of adapting in prevailing circumstances should not interfere with its potential freedom to function in an unknown or uncertain environment. A simple example: human body shows adaptation, while the nervous system exhibits adaptability (e.g. by learning): and not for that we’re saying that the nervous system is an “over the top” on the human body !

So in this evolution we have to change the perspective about the “network”: the network will not be any more an ossified “body” connecting end-points, but it will be an highly flexible communication fabric full of software processing and storage power. As said, it is like bringing it “on the top” (of multiple physical resources, even belonging to other domains). And the value of this “new network” will increase tremendously, as the bet is on its adaptability. A way to monetize ICT technology advances, whilst leveraging one of our major assets: reaching millions of Users.

One may ask: then, where’s the “intelligence” ? Inside or outside the network ? Well, reading Michael Conrad’s book “Adaptability”, the answer is easy. He introduced the concept of “hierarchical compensation”, which is the idea that adaptation predominates at one level and adaptability predominates at another level, in a delicate balance. Needless to be more explicit on the biz implications of this which appears to be natural law.

For sure, the dynamics will be in the hands of Users (who we’re reaching) , to whom we have to look at, not as network end-points but as “minds” playing with the adaptability of this network fabric.

In this new game, there are huge opportunities for Players like us.

Today communications networks include a range of deployed middle-boxes such as WAN optimizers, NAT, performance-enhancing-proxies, intrusion detection and prevention systems, any sort of firewalls, and other application-specific gateways… Each middle-box typically (closed and quite expensive) supports a narrow specialized function (layer 4 or higher) and it is mostly built on a specific hardware platform.

Middle-boxes are deployed along most paths from sources to destinations: that’s why networks lost the initial end-to-end principle of Internet (with packets being just forwarded). In this paper they have presented a measurement study conducted from 142 networks in 24 countries, including cellular, WiFi and wired networks, public and private networks, residential, commercial and academic networks: it sounds incredible, but it resulted that about 33% of paths tested keep state and perform some level of L4+ functionality. Not only said stateful middle-boxes are breaking the end-to-end principle, but they are representing a significant fraction of network capital and operational expenses (due to management complexities).  Above figure is extracted by this paper: it shows that the number of middleboxes is on par with the number of routers in a network. Why do not we get rid of them?

We have mentioned several times that technology records and cost reductions are progressively moving an incredible amount of processing, storage and networking capabilities at the edge. Imagine, as an Operator, to take full advantage of this trend by virtualizing those functions which are provided today by middle-boxes: in other words “dematerialized” and move them at the edge, as much as possible nearer the Users (or maybe some function could be even moved in the Data Centres). The infrastructure will be completely reshaped: Core Network will become stateless (as Internet in former days) and the Edge Networks (and the Data Centres) will become the only stateful parts of the networks! By the way, this is my personal take about the real innovation of Software Defined Networks and Network Function Virtualization: moving the stateful functions of the infrastructure part at the edge and part in the Operators’ Data Centres.

As pointed out by Roberto, in the next decade the edges are likely to be autonomous systems; I’d even say that edges will become the main stateful parts of the network, thus creating a massive distributed data base around the Users. No doubts that the amazing increase of smart nodes and devices at the edges will provide globally enough processing power, data storage capacity and communications bandwidth to achieve this vision earlier than one might expect!

And this, I guess, would allow Operators to re-shuffle the cards of the biz game, even with the OTTs, as money flows in at the first mile: the ability of managing and orchestrating the “states” at the edges, strategically around the Users and across their Data Centres, will create new biz ecosystems with new rules of cooperation-competition between a galaxy of unexpected Players.

Today, we are witnessing the transformation of the edge into a communications environment with high performance computing capabilities, pervasively interconnected by embedded communications. At the edges we are going to see  a growing number of simple nodes and devices powerful enough to collapse all the layers of the ISO OSI stack: eventually all network services and functions (up to applications) will be implemented in software running on cheap general purpose hardware. This transformation will turn the edge in an business arena structured in multiple interacting sub-networks, domains (operated by diverse private, public Players but even Communities). This will boost a galaxy of new ecosystems.

This evolution, enabled by technology progresses and costs reductions, is progressively transforming roles in the traditional ecosystems around the current Telco-ICT business. Declining costs of computation, communication, and storage are, for example, moving the means of information and entertainment production from a limited number of Companies to hundreds of millions of people around the planet. Also we should not forget the rise of large-scale cooperative efforts under the form of open source s/w and h/w development and production, which might create soon a ripple in the Telco-ICT Vendor markets.

Open source projects do not rely on “ossified” hierarchies to organize s/w and h/w productions: people freely participates, with a variety of motivations, and share contributions without anyone asserting rights to exclude either from the contributions or from the resulting whole. This is a radically new production approach, decentralized, collaborative, and non-proprietary, which could impact several segments of economy. And the results are impressive ! Google and Amazon run their Web servers on the GNU/Linux operating system.

These trends, in turn, are influencing the network transformation itself: imagine a network based on open source functions and services instantiated on open source hardware node and devices. This is (almost) possible already today: high performance, low costs and easy-to-use technology “tools” are empowering people to develop the future Internet !

Overall we’ll face the challenge of “complexity”, but I think that this “complexity” itself will bring great business opportunities for Operators.

We know that a throughput of a router is mainly limited by the routing processing, which is impacting the maximum number of packets that the router can process at each time: as a consequence there is an inevitable tradeoff between the number of ports (node degree) and speed of each port (bandwidth per connection) of  a router. Router Vendors cannot make a router that has both a large degree and a large bandwidth per connection mainly due to the limitation of the routing processing.

 Normally nodes in the core network have large bandwidth per connection, and thus small degree, and vice versa for the edge: typically the degree of an edge router is almost five times larger than the one of a core router.

 On the other hand, consider that processing technology advances will make possible (very soon) to build a 100 (or even more) Gbps software router. Or, simply software router architectures capable of parallelizing routing functionality both across multiple servers and across multiple cores within a single server (e.g. RouteBricks). It will be possible to build high-speed software routers using low-cost, commodity hardware. This means that it will be possible overcoming routing processing limitation by using the huge amount of processing power made available in large data centres (if you prefer we’ll bring the control plane of the s/w router – separate from the h/w – in the Cloud).

 Imagine an Operator running a virtual network of such powerful s/w routers on a Cloud and using a low cost physical infrastructure (based on standard hardware) for simply forwarding the packets ? This would change – in principle – the (economic) equation of the network: overprovisioning connectivity rather than just overprovisioning bandwidth. Overprovision connectivity pays off better than overprovision capacity: it is possible creating very large numbers of topologies to choose, even almost randomly (like VL2 and Bit Torrent), or programming and controlling the QoS at higher levels. Up today, overprovision connectivity in a network is more expensive than overprovision capacity, but tomorrow the equation may change.

In a data center, we have already overprovisioning of connectivity, but the story is different: network covers a relatively small fraction of the cost, compared to server, electricity and cooling costs. So overprovisioning connectivity makes economic sense (by the way, in data centers, traffic demands are quite volatile and not well understood, so it is strictly necessary to overprovision connectivity; on the other hand, traffic fluctuation on a network is over time rather than space, thus today is mitigated by capacity overprovisioning).

 In principle, that would mean for an Operator building a competitive advantage by developing Virtual Data Centre, where using big data to control the network and to overprovision connectivity in Virtual Networks… on top of a (very) low cost (Opex and Capex) standard h/w infrastructure (with unlimited bandwidth).

A step further ? Integrate the s/w control of IP and Optical Networks…

You can’t have a network less network, can you? May be not, but this is what researchers are doing and …it works. And it is not just research in some crazy labs, it is being done in industrial settings as well. Ford Motor company has started to install WiFi transmitters and receivers in some of its models in 2010 and has recently stated that by 2015 80% of their cars will be WiFi hubs, ready to talk to nearby cars.

Researchers at MIT, Georgetown University and at the National University of Singapore are studying ways to exploit these bounty of wireless nodes moving around. Why not using each car as a network node, a network that is continuously reconfiguring as cars move around. There are issues about the connectivity and when to drop a link that is fading away and switch to a new link. Complex issues, but they are well known since, to a lesser degree of complexity, they have already been tackled for cellular networks.

However, there are also new issues that are related to the fact that this would be a really mobile networks (in a cellular network all the antennas on the poles are quite “fixed”). The strategy for routing and for the optimization of links usage is quite different and has to be worked out dynamically. There is more. Suppose that your average car has quite a bit of storage capacity on board (a quite natural assumption). Then it makes sense to assume that when a car (a passenger) wants to access some data (like the daily newspaper) it is quite probable that such a data is already available in some cars in the surrounding. So, no need to connect to the big Internet, rather let’s look for the right car.

This is also a question researchers are trying to address. What is the best strategy in distributing data/information in a car ensemble so as to maximize the probability that some nearby cars can deliver the information when needed?

It turns out that, according to the MIT … researchers, cars can be modeled through a paradigm of connected clusters by identifying as belonging to a cluster those cars that have some recurrent contact with one another. Within a cluster a car may have strong links with some (it is often within WiFi contact) and weak links with others. The same will apply to links among clusters.

Now this starts to sound like a small world paradigm….!

Indeed, in the future we can see that the cars, from the point of view of forming a network less network may behave like an ecosystem and the same rules of interactions that are played out at that level can be found in these networks.

And, in perspectives, although this is not addressed in the MIT study, it is not difficult to imagine other components in these network less networks: people with their cell phones that all of a sudden become network nodes of a dynamically fleeting network and also homes…. And what about “things”? They also are progressively connected and they also will be able to act as nodes of a networks that exist because all these nodes are there!

In the next decade we will continue to have big (bigger actually) pipes but these will be connecting communications spaces resembling more to fabric than nets (in the sense of fishing nets). As this happens the role of network providers will change significantly. Players like Google and now Facebook are becoming part of the telecommunications “plumbing/piping” infrastructure and they, as the others, will make use of pervasive communications fabrics. Also the role of service providers is likely to change in this scenario. Rather than delivering a service to a terminal services will have to be delivered to an ambient using pipes to reach it and then leveraging on the local communication fabric connecting all the bit transducers available in that space. And there is more!

A single ambient/communication space will actually consist of several instances, one for each user within that ambient. And the actual transformation of bits arriving from the pipes and interacting with local bits “owned” by the ambient will be orchestrated and directed by the instance owner (that is it will depend on the specific user).

A quite different scenario for a ubiquitous internet infrastructure!

Neuroscientists of University College London (UCL) have found that there is a simple pattern modeling the tree-like shape of brain’s neurons. They have shown how a simple computer program connecting points with links as little as possible can produce tree-like shapes very similar to the ones of real neurons. These shapes follow a power law, which is a mathematical law quite common across the natural world, and underlying complex structures. This is the law:

L = (3/4π)^1/3 × V^1/3n^2/3

where n is the number of dendritic sections to make up the tree, L is the total length of these sections, and V is the total volume

Neuron shape model: target points (red) distributed in a spherical volume and connected to optimize wiring in a tree (black) (credit: H. Cuntz et al./PNAS)

Similar theories about neurons networks have been already published in the past. This time, UCL Neuroscientists tested this theory by examining neurons in the olfactory bulb, a part of the brain where new brain cells are constantly being formed, and found that the growth of these neurons indeed also follows the power law, providing further evidence to support the theory. “The ultimate goal is to understand how the impenetrable neural jungle can give rise to the complexity of behavior” said the UCL Neuroscientists.

Why these results might be very interesting for us ?

Many communication and social networks have power-law link distributions, containing a few nodes that have a very high degree and many with low degree. The high connectivity nodes play the important role of hubs in communication and networking, a fact that can be exploited when designing efficient search algorithms. It has been shown that the Internet backbone and web page hyperlinks have a power-law distributions. And the same distributions might be applicable to future edge networks.

In fact, imagine edge networks evolving towards ensembles of huge numbers of interacting lightweight nodes capable of abstracting communications, processing and storage resources. Millions of nodes, like neurons, embedding simple “hard-wired rules”, will be capable of interacting, self-adapting and self-adjusting to cope with dynamic contexts (e.g. Users’ requests and business goals).

This is very much similar to what’s happening in the brain networks…

In previous posts, we’ve elaborated how a network can be seen as an emergent property of a complex ecosystem. Networks emergence is generally describes as macro-scale properties resulting from micro-scale rules. In the prior-art analysis of network layering, we often find that layers networks can be defined as optimizers maximizing specific composed/aggregated utility functions. If we integrate these two perspectives, in a broader sense, we can define networks as emergent property of sets of optimizers gaming to maximize specific utility functions.

I can see this for natural ecosystems. Imagine a swarm of fishes or even better a nest of ants: the “communication” networks between ants is an emerging property (producing self-organization) to optimize both the “life” of any single ant (which may have its own local utility function) and the organization overall community (having an aggregated utility function for the nest). Actually, I’ve never heard about any risk of “Tragedy of Commons” in ants’ nets: evolution has selected those autonomic behaviors to avoid “breakdowns”, by keeping a delicate equilibrium of local vs global utility functions. A lesson from Nature on self-organization to be learnt.

Emerging collective behaviour (in a swarm of fishes) from simple micro-scale rules

Can we approach future networks, as ecosystems of resources, in the same way ? Which are the utility functions we can design for controlling a communication network ? One can image functions about sources rates, useful information, delay, energy consumption…etc.

In this direction, F. Kelly, has shown that TCP/IP protocol is a perfect example of optimizer: its objective is to maximize the sum of source utilities (as functions of rates) with constraints on resources. And actually, each variant of congestion control protocol can be seen as a distributed algorithm maximizing a particular utility function. The exact shape of the utility function can be reverse engineered from the given protocol. Similarly, Border Gateway Protocols (BGPs) can be seen as a solution to the Stable Path Problem, and contention-based Medium Access Control (MAC) protocols as a game-theoretic selfish utility maximization. Other utility functions could be User satisfaction (e.g. User-generated pricing following end-to-end principle), resource allocation efficiency or different network economics fairness.

Then, modeling networks as emergent property of a sets of optimizers, means considering management-control based on interacting controllers maximizing a combination (e.g. weighted sum) or an aggregation (e.g. in multiplicative form) of several utility functions. Or, alternatively, we may say that management-control should look for the network Pareto optimality, or it should play an uncooperative dynamic game. In any case, this would imply looking at future networks management-control with a different perspective, through the glasses of a deep vertical and horizontal network decomposition.

The emerging paradigm of Software Defined Network (SDN) is about having a fully decoupled network control plane, so it can be seen from this broader perspective, at least at the edge of current infrastructures. In a SDN, control intelligence is (logically) centralized in software-based controllers. Said controllers provide visibility and control over the network, they can ensure that access control, routing, traffic engineering, QoS, security, and other policies are enforced consistently across the network infrastructures. Governing the interactions of these controllers would allow managing and optimizing a SDN according to certain policies, or utility functions.

In other words, approaches like SDN seems paving the way to look at future networks in a different way, as ecosystems of resources, where top down governance (e.g. playing the role of evolution ?) of sets of controllers could meet emergent properties from local bottom-up autonomic behaviors (e.g. via local utilities).

But let’s go even beyond this (partly) engineered approach: can we build only on bottom-up emergent properties (just based on micro-scale rules) to get self-stabilizing future networks ecosystems, indeed like in Nature ? A great challenge towards 0-Capex, 0-Opex networks.

Information permeates everything: from electrochemical information exchanged in networks of neurons, to biological information stored, and processed in living cells, to business information, etc.

Our current understanding of information communication is still based on Claude Shannon’s seminal work in 1948 resulting in a general mathematical theory for reliable communication in the presence of noise..

Claude Shannon

Frederick P. Brooks, Jr., wrote in “The Great Challenges for Half Century Old Computer Science”: “Shannon performed an inestimable service by giving us a definition of Information and a metric for Information as communicated from place to place. We have no theory however that gives us a metric for the Information embodied in structure. . .”

Traditional information theory considers the communication studying the capacity of channels connecting two endpoints. This approach should be enhance when considering  wireless networks (e.g. for example see the posts on Edge Networks) where nodes which relay information in a multi-hop manner and time-varying topology.

In this direction, interestingly, this paper introduces the concept of the spatio-temporal relaying: information is carried from a mobile transmitter (space) in its past (time) to a mobile receiver (space) in its future (space). Nodes that forms a path in a spatio-temporal space of information transfer: the quality of the transmission depends on the respective spatio-temporal positions of the transmitter and receiver. So a grand challenge is to extend Shannon capacity formula to multi-source wireless networks.

This may have impactful applications: recent researches on MANETs has led to definition of the so-call “space-time capacity paradoxes”. Theoretically, the capacity of a multi-hop wireless network increases with node density and node mobility in spite of the apparently effect of transmission interference.

Moreover, it has been shown that the theoretical capacity of a multihop wireless network is proportional to the square root of the network size (number of nodes). This  promises enormous wireless capacity for ultra-dense networks ! On the other hand if you try testing this on WiFi networks, capacity has a tendency to decrease with the number of nodes, rather than increase as theoretically predicted. This reflects the fact that the WiFi medium access protocol, primarily designed for wireless LANs, does not scale to multihop networks. A breakthrough seems to be possible here.

In these areas of study, National Science Foundation has established the Science and Technology Center for Science of Information to advance science and technology through a new quantitative understanding of the representation, communication and processing of information in biological, physical, social and engineering systems.

Network are becoming more and more complex and dynamic, capable of interconnecting large numbers of resources (e.g., routers, switches, transport nodes, servers…), Users’ devices (e.g., smart phones, etc) and, in the future, any machines (e.g. sensors, smart things, etc) embedding communication capabilities.

Future networks will be similar to complex systems where global properties and effects can emerge abruptly at a critical level of interactions between their components. In these dynamics, there is the hidden risk of instabilities. Overall, instability may have primary effects both jeopardizing the network performance  and compromising an optimized use of resources. In the worst case, an instability may create even a meltdown of a portion of network.

This is the main problem which I’ve proposed for study (more or less on year ago) in the EU project Univerself as part of the Telecom Italia participation in the project. Basically, we’re looking for methods and systems able to ensure network stability through local self-adaptation of nodes and, if-when not sufficient, via centralized policy based control.

This morning I’ve been very pleased to read this interesting paper Icebergs in the Clouds: the Other Risks of Cloud Computing addressing the risk of instabilities on the Cloud, which is essentially a metaphor for a network of computing and storage entities in which tasks and resources can be shared.

Example instability risk from unintended coupling of independently developed reactive controllers

Paper points out that complex systems can fail in many unexpected ways and outlines various simple scenarios. In the worst case, a cloud could experience a full meltdown that could seriously threaten any business that relies on it. Well, this is very much the same for future networks!

A growing number of researchers are beginning to see this problem: unpredictable behaviors often emerges in systems made up of “networks of networks”.

Paper concludes with the following: “We should study [these unrecognised risks] before our socioeconomic fabric becomes inextricably dependent on a convenient but potentially unstable computing model.”

Neurotransmitters are chemicals produced by the nervous systems in order to relay a nerve impulse from one cell to another cell. In the brain, neurotransmitters have a central role in shaping memory, learning, mood, behaviors, sleep, pain perception, etc. Basically they operate at the junctions between neurons, allowing communications: when an impulse arrives at the end of an axon, neurotransmitters are released, diffusing across a gap to the next neuron; each neurotransmitter binds only to specific receptors on the postsynaptic membrane.

There are many types of chemicals that act as neurotransmitters. For example, serotonin plays a major role in emotions and judgment, and also sleep. Endorphins are neurotransmitters that relieve pain and induce euphoria.

Neurons interconnections and Neurotransmitters

So brain self-organization is determined basically by two main phenomena: local reactions (firing) of neurons, due to the exchange of electrical signals through neurons’ interconnections, and the global influence of the neurotransmitters.

Imagine taking this picture for managing future networks. As it made no sense in Nature managing or controlling the behavior of a neuron, in the same way we should not expect a centralized management in charge of the hundreds of billions of electronic devices, machines, smart things connected with each other and to the Internet (but on the other hand, we dream to have the Net well self-organized as a … brain).

Therefore, learning from Nature, let’s imagine future network nodes capable of local reactions to the context (as neurons do through their interconnections) and then a global harmonization (as neurotransmitters do) of all these local reactions through the viral propagation of context information (a sort of reaction-diffusion process). Can you see it ?

 In a next post, I’ll make a proposal for a concrete proof-of-concept with today technologies.

Let’s look at networks with different eyes to “simplify” the future !