Yesterday Antonio wrote a nice post on the growth of capacity, processing and storage, at the edges, drawing some interesting implications.

Assembling a Cray 1 back in 1976.

That made me think about “the speed” at which this capacity is actually growing…

The recent announcement of the Mac Pro, just last week, was an clear confirmation of the amazing growth that is still taking place. I don’t know about you but the shape of the Mac Pro makes me go back to the Cray 1, built in 1976, the fastest computer in the world! It had an incredible processing power of 100 MFLOPs, and its successor the Cray 2, in 1985 reached 1.2 GFLOPS.

Well, compare this with the 7+ TFLOPS delivered by the Mac Pro. The Mac runs 70 thousands times faster, consumes 1,000 times less power, has a size that is 1.700 times smaller and the cost… well, it has not been announced but it is surely going to be less than the 10 M$ that was charged in 1976 to buy a Cray 1!

So this is the kind of progress the computer industry has made in 37 years! You can say that this is less that you would expect from the Moore’s law: according to Moore’s law one should have achieved an increase of 70 thousands times in “just” 25 years, not in 37 but wait. Here we are comparing apple and oranges. We are comparing a supercomputer (Cray 1) with a mass market (almost) computer. Here we are saying that the performances are 70,000 times better AND the cost is 1,000 times lower! That is a factor of 70 million and that is EXACTLY in line with the Moore’s law! As a bonus, you get a form factor that can let you place the Mac Pro on your desk and you can power it with a fraction of the power used by your fridge. On the contrary, the Cray 1 needed many “fridges” to air condition the room.

The new Mac Pro, 6.6 inches per 9.9 inches, packaging supercomputer power…

Comparing the Cray 1 with today’s Thiane-2 (announced just few days ago, designed and built in China has a performance of 33.86 PFLOS, that is 330 million times faster than the Cray 1, and that is actually faster than what Moore’s predicted by a factor of 10. Indeed experts were not expecting this performance before 2015, we are witnessing an acceleration rather than a slowing down of the Moore’s law!).

The dramatic reduction is cost, making a Cray 1 affordable to all people, as Antonio pointed out a smart phone today has a processing power that is twice as much as the one of Cray 1, has really created huge processing capacity at the edges of the network. Indeed, we should notice that as Moore’s continues to push increased capacity on chips, the reduction is cost multiply the dissemination of those chips, resulting in a capacity that is actually increasing faster than what is forecasted by the Moore’s law.

So far we have not learnt to exploit it as a whole. We are using it to run apps, to have better video but most of this capacity, differently from what happened in the Cray 1, is simple wasted, it does not get used. We are exploiting the increased performances of the single chip, not the one deriving from all the chips taken together, and we have just said that this  capacity is increasing at a faster pace than the one of the chip.

This growth of capacity at the edges will change the way services are designed, and operated. It will displace the Telecom Operators business and open the door to a multitude of service providers (to some extent it already has).

New architectures are needed to exploit this untapped capacity. Does it make sense do work on this? Well, from an economic standpoint, the processing power is so cheap that making smart use of it has really low economic incentive.

On the other hand, that unlimited capacity at the edges may open the door to new applications, based on a crowdsourcing model. Parallel evolution in sensors with massive, distributed deployment, in halo nets are likely to take advantage of this capacity. Low power consumption and faster recharging are going to tip the architecture of networks and services in the coming years.

Share

• 

Today a smart phone has a processing power of about 200 megaflops, a laptop is offering some tenth of gigaflops, a PlayStation hundreds of gigaflops. Imagine to find the way for orchestrating millions of said Users’ devices, harnessing their idle processing and storage power: we can achieve bigger capacity than a supercomputer, like Titan (today number one, capable of 18 petaflops).

This distributed platform of edge devices can indeed create a sort of processing and storage fabric that can be used to execute any network function and to provide any sort of ICT services and applications. The components of this fabric can be seen as: CPU/GPU, SSD (Solid State Drive), HDD (Hard Disk Drive) and link (and this is perfectly in line with the “disaggregation of resources” targeted by the Open Compute Project).

One may imagine these components aggregating dynamically in an application-driven “flocking”. And, in the same way as birds with simple local behaviors are optimizing the aerodynamics of the flock (which is solving a “constraints optimization problems” by using very simple local rules), the flocking of component can follow dynamically application-driven network optimizations.

The problem is finding these local rules. Not only, but also the optimal way to allocate and dynamically migrate Virtual Machines and data (which are representing also states). Let me make an example, a very simple model. Imagine, just for didactical reasons,  to consider the equivalence between the time of one CPU cycle and the time of a step in a walk. The latency in accessing a SSM (e.g., DRAMs) can be estimated as around tenths of CPU cycle, tenths of steps in our example. But if you wish estimating the latency in accessing the HDD, i.e. the stored data (also including the latency of the network links, RTT), then overall it results the time to make a walk of about 10 000 km.

I’m sure that solving this constraints optimization problem…will mean allocating processing and storing data as closer as possible to the Users!

Share

• 

Researchers at the Urbana Illinois University have developed a cradle and an app for the iPhone transforming it into a Bio Lab.

The iPhone becomes a bio sensor. Photo by
Brian T. Cunningham

The cradle and app are using the iPhone camera and processing capability transforming it into a biosensor to detect toxins, proteins, bacteria, viruses and other molecules.

The bio sensing is made possible by a photonic crystal. This crystal lets pass just on wavelength  so that when anything biological attaches to the photonic crystal the reflected wavelength will be an indication of the bio substance that has attached to the crystal. This bio substance can be a protein, a cell, a bacteria, a virus.

The bio sensor cannot detect “any” bio molecule, it has to be primed to react to a specific target. For that a normal microscope slide is coated with photonic material that can “detect” the desired molecule.

This photonic crystal slide is inserted in the cradle and the iPhone camera (the app using the camera) measures the spectrum. The reflected wavelength shows up as a black gap in the spectrum. Then the measure is repeated without the microscope slide inserted and the degree of shift in the reflected wavelength indicates the amount of target molecule in the sample.
The resulting bio-sensing device is not as precise as a lab measurement but it can provide information that is not available in the lab: one can measure in real time and in several places the presence of a molecule, a toxin, using the iPhone GPS to localise the sample and this can give a good indication of the spreading, as well as the origin of a given bio molecule.

Clearly this is not for the every day user. It is a tool for bio specialists. But over time I can see that more and more sensors will become available for smart phones making detection of certain substances an everyday experience for everyone of us. We would actually refer to our phone, and be warned by it, for detecting allergenic substances or specific bacteria that may be dangerous given a specific health care problem we can suffer from.

Share

• 

Live cockroaches with a controlling backpack receiving signals for a remote and interacting with their antennae to steer their movements

I have published, some years ago and again few months ago, the news of scientists trying (and succeeding) to interact with cockroaches by sending signals to their antennas and therefore forcing them to move in a certain direction. The research has two objectives: a better understanding of how the brain works (and yes, neurones in a cockroach brain are exactly like the ones in our brain…) and exploiting the capability of a cockroach to move under rubbles to find victims of an earthquake buried under a collapsed building.

I have to say that I felt a bit of sympathy for the cockroach but the goals seemed to me to be worth the inconvenience (pain?) brought to the animal.

But now, I run into a project seeking funds on Kickstarter that aims at transforming a roach into a sort of micro-machine remotely controlled by a cell phone. The proposers are claiming that the cyber-roach will help kids understanding neuroscience.

The idea is to sell a kit containing a chip (as the one shown in the figure above) and some needles that can be inserted in the antennae of a roach (or in its legs) and wirelessly connected to a smartphone.

The kit comes with an app through which it is possible to interact with the insect and steer its movement as if it were a micro-machine.

Controlling a roach with an app on your smart phone.

Needles in its legs can detect the spikes created by brain commands and are relayed to the smart phone and displayed on the screen so that one can see what is going on as the animals move around.

According to the proposers the kids using this device show a better grasp of neurology and that demonstrate the usefulness of their product.

They are also showing a video clip on the Kickstarter web site as they advertise their idea and ask for funding. You can see it through the link I included. I decided not to include it here because I really don’t like using a roach as if it were a micro-machine. My sympathy is all to the roach, none to the proposers.

Nevertheless, I have to admit that we are more and more in the understanding of how our brain works at a “mechanical” level and we are learning how to interact with it, with tools that are easier and easier to use and affordable. This is going to create a nightmare in the next years. Will the wedding ring eventually contain some sort of interface letting our partner to influence our life at a “mechanical level”, would it open up a window on our thoughts? The scary aspect is that this is no more science fiction, just science….

Share

• 

Once in a while I think it is better to look backward and see if there are hints we can find in the past that point us to possible futures…

Yesterday I gave a talk on the possible evolution of NGN, Next Generation Network, and since I gave the talk in Milan to a bunch of foreign people I decided to start with the Milan Cathedral.

The Milan Cathedral, the heart of Milan.

It is a magnificent monument whose history goes back to the XIV century. What could be farther away from an NGN than this cathedral. Well, consider the following:

• it was not built from scratch, on a green field. It was built over at the place where there was the St. Ambrose church and in turn that church was built in the V century over the old remains of a IV century church (Battistero Paleocristiano). The builders of the cathedral had to create “interconnections” with these previous “infrastructures” (and today as you visit the Cathedral you can still visit the remains of the previous churches).
• the construction started and restarted several times, each time to be stopped because something (in most cases the lack of funding) required a reconsideration of the plan. And the restart brought in new technology that led to some slight revision of the plan. Over time it resulted in a quite different outcome that would not be recognised by the original architects.
• To manage the work a new company was created, FABBRICA. We are still using that word in Italian for “factory”. It was felt, at that time – that is 1387- that an independent company was needed to take responsibility for such a tremendous endeavour. The Cathedral is, still today, the 5th largest one in the world. Just few days ago a decision was taken by Telecom Italia to spin off the distribution network as a first step to create a separate company in charge of deploying the NGN.
• The large funds required by the construction were obtained through a mixture of private and public capitals, what we call today a PPP: Private Public Partnership. Exactly the approach under discussion nowadays in many Countries for the deployment of the NGN.
• It was soon clear that the existing regulatory framework (that is the way of paying a fee for any construction material as it was transported across various townships and provinces) was not sustainable. Hence a special law was approved to change such framework for the construction of the Cathedral. Stones, wooden beams and marble became tax exempted and the pallets had the sign: A.U.F. (Ad Usum Fabricae – to be used by Fabbrica). We still have a remnant of that in a way of saying in Italian: “a ufo”, that is to do something (like traveling, eating…) without paying.
• To make money a variety of “biz models” were implemented, including “advertisement”. You could have a bench in the Cathedral with your name, or have your name engraved in a column ….
• It took almost 6 centuries to finish the construction: from 1386 to 1965! Let’s hope the deployment of the NGN won’t take so long! Don’t laugh! I wouldn’t be surprised if indeed the completion of the NGN deployment would take more than that: you need to consider that the Duomo construction took 579 years but we are living today in an exponential evolution space. That is, in the next 18 months we are going to see the same evolution that took place in the last 50 years. If you do the math, you’ll discover that 579 years are today compressed in less than 14 years and, well, are you ready to bet the the full deployment of the NGN will take less than that? And now that I thinks about it: in Italy we started to plan for the NGN (and we started deployment of coax cable with the Socrate project) in the last century (the first deployment of Socrate was in 1995!) and since then we have had a number of stop and go with rethinking of the NGN architecture (actually the name NGN came later, in the middle of the last decade). Discussion on FTTH, FTTC, GPON is still going on, so technologically speaking the NGN has already taken more than the construction of the Milan Cathedral and it is far from complete. As we say, so far the only fully deployed architecture in NGN has been the FTTP, Fibre To The Press…
• In the end the Cathedral was completed and it stands as a masterpiece benefitting the whole city of Milan, attracting business and generating wealth that does not go to the FABBRICA (it still exists, under the name Veneranda FABBRICA del Duomo) but to many OTT (that’s how we call them , don’t we?).

Share

• 

Open Compute Project (OCP) is an initiative started (on April 2011) at Facebook with the goal of developing “one of the most efficient computing infrastructures at the lowest possible cost”. The project is pursuing an open hardware approach, to “develop servers and data centers following the model traditionally associated with open source software projects”. In other words, the project is aiming at disaggregating the servers of data centers into elementary sub-systems that can be swapped out depending on applications’ needs. Memory, network, and storage resources would be disaggregated and shared across the rack and fast interconnects (e.g., 100 Gbps) will provide the binding of these elementary sub-systems.

And next target of OCP appears to be “disaggregating the network”. Basic idea is developing network switches that will look like more disaggregated open servers. Then, instead of having a vendor-specific network OS, one may want to load with any trusted software provided by developers. This OS-agnostic disaggregated switch is expected to enable a faster innovation in the development of networking hardware as well as costs drops in infrastructures exploitations; industries like Intel, Broadcom, VMware are already moving in this direction, also in the context of other similar initiatives.

Technology advances exploited in initiatives like this are concrete milestone in the ICT Fabrics vision we’ve elaborate some posts ago and this paper: imagine disaggregating (from a functional viewpoint) processing, storage and networking at more granular component and then composing, orchestrating them (beyond the borders between date centers and networks) through fast optical interconnections.

As Roberto mentioned in the post Telecommunications as a gigantic supercomputer this is not just a speculation, but it is likely to become a real medium-long term r-evolution capable of enable new networks and new services. The edges (around the Users) will look like virtual Data Centers morphing dynamically in space and time and capable of storing big data and information and the “flocking” of resources component will be “application-driven” without any networks-DCs borders.

Share

• 

As I pointed out in the previous two posts, as human beings we need much more bandwidth than the one we get today in most parts of the world. From a technical point of view, even the lowest capacity infrastructure we can deploy today (it is already being deployed) is more than enough to meet our needs. Unfortunately the problem is not lack of technology but lack of money and lack of culture.

Credit: FCC

In some parts of the world there is confidence that ultra broadband, over the 100 Mbps (and I would like to set the need at 500 Mbps downstream and 200 mbps Upstream) is already a defined roadmap, as shown by the graph here forecasting UBB penetration in the US.

In many parts of the world bringing that much bandwidth will be a dream for at least two decades if we think about using optical fibre (as it is the case in the US depicted in the graph). Developing economies have seen an amazing progress in terms of telecommunication penetration (not just China and India, also Africa is moving towards a 100% penetration of cell phones) but the bandwidth provided is clearly limited.

A bandwidth of a few Mbps is within reach almost everywhere using wireless communications (it may take this all decade to get there) but moving from a few Mbps (even with FRA – Future Radio Access) to hundreds of Mbps made available to all homes is beyond the economical possibility of most Countries. That sort of bandwidth requires pervasive fibre infrastructure (with radio drops in cells that are no more than 100 m, possibly smaller).

Alternatively, it requires completely different paradigms that today are just a matter of speculation.

In a number of Countries, like most of Western Europe, a pervasive fibre infrastructure is economically possible, but it is not able to generate that kind of ROI (Return Over Investment) that stimulates private investment. Private Public Partnership in these cases is almost mandatory. However, even a PPP is not enough unless there is a commitment to leverage the UBB, by re-engineering the Society processes (eGovernment, eHealth, eEducation, eLogistics, eProduction….).
UBB can increase the efficiency at Country level, estimates are in the order of 5-8% corresponding to a huge return of investment at Country level, able to recoup the investment in ONE YEAR! However, the destruction of “inefficiencies” results in destruction of jobs, and companies that are today living out of this inefficiency. And the corresponding creation of new jobs is not in synch with the destruction…

Hence, the risk is that the PPP investment does not result in an increased efficiency but in an further burden on the Society, increasing the cost.

Share

• 

Having seen the theoretical bandwidth consumption capability of our senses (using grossly simplified measures but that is ok for the point I want to make…) we have to realize that our senses do not consume bits…. they consume images, sounds, fragrances, acceleration and so on… Hence we need to convert bits carried by the network into these images, sounds…

A GPON infrastructure can sustain the bandwidth matching our sense’s resolution

Since we have seen that what gets the lion’s share, in terms of bandwidth, are the eyes we can focus our attention on them.And the translators that we use to convert bits into images are called… “screens”. Today we have HD screens (HDTV) that can display 2 Mpixels and require 10 Mbps bandwidth. The first 4k screens are becoming available, although we can expect them in full swing by 2015; they have 8 Mpixel resolution (like our eyes) and require 70 Mbps bandwidth. By the end of this decade we might see the first mass market 8k screens with a resolution of 32 Mpixels, exceeding our “brain” vision capability and providing a full immersion sensation: that will require a bandwidth of 150 Mbps.

Of course that goes for a single user. In a home you have several users although it is likely that the shift towards higher resolution screens will be gradual: only one 4k screen in the home for a while and then, some 5-7 years later only one 8k screen. Actually, 8k screens are not likely to become pervasive in the home, there will be probably just one, since it only makes sense if it is really big (over 80″, better over 100″) and you don’t have that much space in a normal home. You are likely to have, as it is common today, several screens but they will be HD or lower resolution, a few (2-3?) 4k and one 8k. Hence, it makes sense to assume as an upper boundary for bandwidth requirement at home something like 300 Mbps.

Of course, in other ambient, the bandwidth demand can and will be greater but in terms of economics the mass market is the one leading (for both cost and revenues).

What we can see is the the lowest bandwidth fibre infrastructure, the GPON is and will be providing more capacity than what is demanded. Hence it is a future proof infrastructure.

Share

• 

I have been often confronted with the question: How much bandwidth do we need? The straightforward question of course is: As much as I can get, no boundary to my wishes…

On the other hand, I realise that beyond a certain bandwidth I may no longer be prepared to pay for more and that, from an investor point of view, places a limit to the bandwidth that makes sense to provide.

Rough estimate of human senses resolution in terms of bits, based on good compression schemes

Let’s consider our senses from the point of view of how many bits they can be considered equivalent. Obviously, the point is that a sensor able to capture a maximum of 10 bits per second would be completely insensitive to receiving more bits, which in other words means that those other bits are useless in terms of transmitting a message, hence no point in sending those bits…

Our eyes retina has a resolution of about 8 Mpixels (like a medium digital camera, although along with the software, as we may call it, processing the detected image through three layers of cells, the eye has a much better latitude and it also has some very good capability to detect edges and movements -these latter are becoming features of the newest digital cameras as well). Having two eyes one might say that we have a reception capability equivalent to 16 Mpixels but that is not true since the two retinas are capturing almost the same image using the slight difference due to the distance between the two eyes to send signals that create the 3D perception in the brain). Overall the two eyes have a total resolution that is in the order of 8 Mpixels but we might want to take into account the need for sending two different (although very similar) images and therefore from a computation point of view we should count an equivalent of 16 Mpixels which requires a bandwidth between 100 and 200 Mbps (depending on the type of cmpression that we can adopt and some sophisticated mechanisms for reducing the number of bits sent). Notice that the bandwidth required depends on the refresh time, i.e. on the number of images that the eyes can capture every second, between 16 and 20 per second.

Our eyes, however, are in continuous motion, saccadic movements, and as a matter of fact they cover a larger area that is equivalent to 28 Mpixels. This area is constantly refreshed through the saccadic movements and resides not in the eyes but in the brain, in the visual cortex. All in all we have then to take into account these 28Mpixels, which means using a bandwidth in the order of 400 Mpixels.

All other senses use a much more limited amount of bits, and a corresponding lower amount of bandwidth.  The second sense, in terms of bits detected is the touch that although may require up to 1 Gbps because of the very low latency (1/1000 of a second) in practice works per segments and cannot manage more than 100 Mbps. The bandwidth is therefore just one fourth than the one absorbed by sight. However, it poses greater constraints in terms of latency, since it has a “refresh rate” of 1/1000 of a second versus the 1/16 required by the eyes.

Our other senses, hearing (400kbps), nose (40 kbps)  and proprioceptors (100 kbps) are 2-3 orders of magnitude less bandwidth hungry, so we can basically disregard them and focus on sight and touch. Notice how all these figures have been derived by multiplying the number of individual sense termination (cone, rods in the eye, hearing nerves, pressure sensitive cells and so on) with the maximum reaction frequency (any sense cell can only fire as many times per second and therefore can only send as many signals per second, each signal being a “1″, its absence being a “0″).  Also notice that this flow of “bits” may not reach the brain, actually, most of it does not reach the brain. Signal integration occurs at several levels, within the retinal layers themselves, at the spinal cord (for touch signals…) and so on. The only signals that are NOT preprocessed before getting to the brain are the olfactory cells that as a matter of fact are nerve terminations directly stemming from the brain.

The real flow of signals reaching the brain at any instant is estimated in around 2 Mbps, but in order to generate such a flow our sensory system needs to be stimulated with a bandwidth in the order of 500 Mbps. Anything above that is simply not leading to any new info from the brain standpoint.

Notice, however, that our perception of the world involves a much complex “bandwidth” within the brain. The 2 Mbps that are considered as the result of preprocessing and that are reaching various parts of the brain are further (in engineering terms…) collapsed into very few semantic bits making us perceive “a dog”. That message can be the  result of bits deriving from the computation of an image, from the computation of a sound, from the computation of feeling a furry texture… and so on. At the same time, those semantic bits are not “a signal” rather a complex state where memory, processing and “energy” are all involved. Although the bits may say “dog” the attached feeling may involve pleasure, fear, perception of a real dog or of an image of a dog or even of a fake dog, like the one that may roam in a fable….

There is actually very little ground to day to quantify these semantic bits, since as I said, they do not refer to a signal, like the bits that we pretend arrive from our eyes or ears. They involve billions of neurones, how many we really do not know today. Most (all?) all these neurones are also engaged in other semantic bits and differences may be in the state/energy activation levels of each.

Anyhow, from the point of view of gauging the required bandwidth as long as we are relying on senses this internal brain processing can be disregarded.

Share

• 

At the South Korean Jang-Ung Park Research Centre a team of researchers have developed a soft contact lens embedding a single pixel diode and a sensor able to analyse the eye “tears” and if a problem is detected the pixel lights up setting up a red flag that is seen by the wearer of the contact lens.

Researchers embedded a light-emitting diode into this contact lens. Credits: Jang-Ung Park Research

Apparently, the tears that are “wetting” our eyes contain a number of substances that can provide clues on the health state of that person.

This of course is a long term objective, provide a monitoring system that you can wear on your eyes, able to signal in the most seamless way that there might be problem. Researchers believe that moving from one pixel to several, thus providing more information display capability…, and at the same time provide more sensing capability and more “processing” capability. Powering may come from the chemical energy contained in tears or from light reaching the eye… Clearly, putting all this things together is going to take all of this decade so that we are not going to have this capability of looking at our “health” before next decade (although the researchers are convinced that much sooner we might start to see applications for monitoring specific issues, like high pressure for glaucoma…).

The first stepping stone, however, has been laid. The researchers have found a way to embed electronics in a normal soft contact lens by using two layers of graphene sandwiching silver wires. The dimension of each individual component is so tiny, measured in nanometers, that they remain completely transparent. One of the challenges was to layer the electronics at low temperature (normal silicon based electronics require high temperatures that would destroy a contact lens). For this they have found a method of creating the circuits using a liquid deposition.

Share

•