Lawrence Roberts: Network intelligence for UBB

The Internet has many fathers, but none has contributed as much for as long as Lawrence Roberts. In the 1960’s, Roberts created the ARPANET, from which the Internet is a direct descendent. Normally this would be enough for one lifetime, but Roberts hasn’t been content to rest on his laurels. He’s spent his entire professional life trying to guide the development of network technology as he thinks best, and even now at the grand old age of 77, he’s evangelizing his latest proposed upgrade for the Internet – Flow-state Access (FSA).

By Jason Patterson & Linda Xu

The Internet has many fathers, but none has contributed as much for as long as Lawrence Roberts. In the 1960’s, Roberts created the ARPANET, from which the Internet is a direct descendent. Normally this would be enough for one lifetime, but Roberts hasn’t been content to rest on his laurels. He’s spent his entire professional life trying to guide the development of network technology as he thinks best, and even now at the grand old age of 77, he’s evangelizing his latest proposed upgrade for the Internet – Flow-state Access (FSA).

A little history

The notion of a worldwide network of computers is often attributed to a paper that American scientist J.C.R. Licklider wrote in 1961 that proposed what came to be known as the Galactic network (described as such because it’s proposed architecture resembled the eponymous celestial system). Lawrence Roberts was in the midst of his doctoral studies at MIT at the time and this paper had a major influence on him. When asked at the Huawei Ultra-broadband Forum (UBBF) about the effect that this notion of a massive computer network had on him as a young man, he stated, “He (Licklider) convinced me that was an important goal and the next step that needed to be done and I knew how to do it. So I undertook to do it. I believed that we would be able to connect all of the computers in the world, though I believed that it would be only thousands, because we had big mainframes then. What I really believed was that we really needed to make the data in every computer around the world available to everybody so you get all of the data, and we’ve now got that.”

Roberts would find himself within a few years at the U.S. Department of Defense Advanced Research Projects Agency (ARPA), and it was there that he would do his seminal work culminating in the ARPANET in 1969. The ARPANET, and later the internet, was based on packet-switching technology, of which Roberts was one of the creators. At the time of its development, packet switching was seen (and rightly so) as a potential disruptor of the then prevalent circuit-switching technology, and Roberts has spoken in the past about how there was a great deal of resistance at the time to packet switching from colleagues who saw the technology as a threat to their livelihoods. This resistance no doubt delayed the commercialization of packet switching, and perhaps the Internet, by some years.

Quick wins are out there

Today, disruption is much more in vogue, and for laymen it’s hard to imagine something as revolutionary as packet switching being ignored today, but according to Roberts, there’s still low-hanging fruit out there when it comes to the network. In our interview, he gave two examples. One was security, without which industry ambitions for the IoT, IoV, and smart cities, among others, will be stillborn. “Security could be put onto the Internet, not just in the computers. We are losing the security battle on computers. But if the network participated, it could greatly improve security since all remote hacking is over the net. The security, so far, has been totally concentrated on changing the computer. So, you try to block all of the holes in the computer. But the computer has so many holes because there are millions of lines of code in the operating system that it’s almost impossible and we are losing the battle. We’re getting more and more hacking everyday that’s feasible. That can only be changed if the network intelligence is improved to watch for that kind of hacking, and that can be done; it’s been shown that you can detect bots on the network and you can detect their traffic because it is a very different kind of traffic, if they’re hacking. So we can change that. We can improve it, but we have to change the network, which nobody has been willing to do.”

The other example of low-hanging fruit that Roberts gave was FSA. According to Roberts, “The FSA protocols I am speaking about were standardized in 2011 and 2012, but no use outside of DARPA tests have been made yet.”

About FSA

Flow-State Access was developed by Roberts to help overcome some of the Internet’s current limitations, specifically those of the Transport Control Protocol (TCP). As to what those limitations are, and how they hinder the network, Roberts did not mince words. “We’ve done nothing, since I started the network, in terms of improving the capability or anything but the speed of the Internet. Speed is going to increase because of Moore’s Law, so we keep on building new equipment that’s faster, but we haven’t changed the intelligence of the network, and we are still very dumb in terms of how we manage the traffic. What’s limiting the quality today is the fact that TCP was initially designed in the late 70’s. It was based on the speed of the network at that point. And it worked fine, up until about now. Now, if we try to move 4K TV, we can’t. We need to keep on thinking about how the network manages traffic and manages its own control and the concept then was to let the edges control the rate, but that leaves a lot of problems that haven’t been solved at the higher speeds. TCP is rate limited by distance, router queues, and packet discards. It can only be fixed by adding function to the Internet rather than something in every computer. The FSA protocols I have standardized can fix the problems without scrapping TCP, which is everywhere.”

Roberts also sees traffic increasing at Moore’s law (doubling every 20 months) as it keeps the expense He would add, “Moore’s law expands the communications capacity every year and the human response is to send more data in terms of video density, webpage size, file size via email, etc. Since TCP is flow-rate limited and reaching its limits as it stands, we need to improve the network overload control to allow TCP flow rates to expand for higher rates and lower delay.”

With TCP, basically if the flow rate overloads the available bandwidth, packets are discarded (irretrievably), and the transmission rate becomes very slow. With FSA, there is a continuous feedback mechanism that makes that certain that transmission never overloads the available bandwidth, so that the flow is smooth and continuous, with no loss, with the trunk filled to 90% of capacity (if possible).

What can FSA achieve? According to Roberts it achieves, “ultra-low delay, very fast flows, and no loss (discards).” This translates to data streaming that’s 1,000 to 50,000 times faster than TCP. Roberts claims that this will prove more than adequate for 4K/8K streaming, and remote response times on par with what a human could achieve onsite, thus enabling applications like remote surgery, cloud gaming, drone operation, and advanced robotics, even across continents.

What’s more, with SDN now starting to take hold in the telco industry, Roberts sees FSA as fairly easy to implement. “A service provider can just add FSA software to their SDN. There is no need to replace all the billions of dollars of equipment already installed and the QoS and capacity of the network can be fixed over time. It can be done incrementally (using proxies until it is added to endpoints), without a major problem, like we have with IPv6.”

If he could do it again

Speaking of IPv6, when asked what he would do differently if he could design the Internet all over again, Roberts responded. “We should have started with a variable address size. We have had to change protocols several times due to address space. We avoided variable-size addresses because it was deemed too difficult back in the early days. But it soon became easy and we now have the huge expense of converting to IPv6, with no major gain except adding bits to the address space. Adding a byte more of address to IPv4 every decade would have been a much better plan.”

He later added, “I think that if I had continued in my position at ARPA longer, I would have started working on security earlier. But, as it was, I turned it over to the next generation of people at ARPA and security never got implemented so we needed to be working on that earlier in time because now, of course, it’s a big problem. But, we didn’t have a problem in the first decade because everybody was cooperative, but quickly it became so some people weren’t cooperative. So that was one thing we should have changed.”

What it all means

One thing that has not changed is Roberts’ basic view of the Internet as a medium for information access. During his working on the ARPANET, he largely viewed it at the time as a means for research institutions and the like to communicate and collaborate. Even today, he retains this view of information availability as the most significant effect of the Internet. “All the information in the world is available to everybody so that they can find out anything without having to have any other… books available. They don’t have to go to the library. So, they can find out instantly about whatever they don’t understand.”

In the era of social media, this perspective might seem old hat. But don’t be fooled. Roberts’ mind is as focused on the future as any. At the UBBF Forum, he provided a look ahead that he dubbed “work at home, robot at work” – a vision of advanced remote-controlled robotics involving such tasks as home assistance or remote surgery. Such robots might be controlled by man or machine, but in order for a machine to do the work of a man, it needs comparable reaction times of 50 to 100ms at the most. According to Roberts, TCP currently allows about 500 to 2,000ms. A lot can happen in two seconds if a driverless car is traveling at freeway speeds, so we’ll definitely need a leap in network capability on par with FSA if this vision is to come to pass.

But the benefits would be enormous. According to Roberts, “We won’t need quite as much doctor involvement directly (through cloud-powered artificial intelligence) because for many of things, we just want to know what the right treatment is. Education can be changed dramatically as universities offer these courses available worldwide on the Internet. That can start happening in all the educational ranges. You can get an education without having to be at a school.”

He later added, “I think it (improved connectivity) will change the developing world a lot because a cell phone is now relatively cheap. They can get their first step into the digital world that way, even if it’s only a small screen. Of course, the screens are getting bigger on the phones. But, they’ll be able to keep on with relatively inexpensive devices because electronics keep getting cheaper and will be able to get full access to the Internet, and that’s important, because that’s how they are going to educate themselves and grow and found out the information that they need. If you are going to compete in India or China and you live in a rural area where you can’t get the right communication, you can’t know what everybody else has already done. But if you’re on the Internet, you can. And so, very quickly people are doing that, because they need to in order to compete, in order for their contribution to be valuable to the world.”

About Dr. Lawrence Roberts:

Dr. Lawrence Roberts is the telecommunications expert with over 30 years of experience developing, marketing and selling packet networking equipment and service. He is also the founder and CEO of five telecommunications companies. Known as the founder and architect of ARPANET, he is the visionary expert at creating technical breakthroughs, bringing together management and technical teams and introducing innovative products to solve business problems.