william87 - Fotolia
Published: 02 Feb 2015
The Internet was not built to support Netflix. And for a long time, that was perfectly OK.
When the architects of the Internet started interconnecting communications networks, the types of applications they needed to support were simple by today's standards: remote login, email, file transfers and, later, Web access. What became known as the Internet protocol suite, TCP/IP, was well suited for the task.
Almost immediately after the public Internet's birth in the early 1980s, the industry developed the domain name system (DNS) to translate the numeric IP addresses given to endpoints into names like acme.com. The goal was to make it easier for people and applications to understand in order to exchange data. But the reliance on TCP/IP and DNS led to the creation of two separate namespaces -- one for numbers and another for names -- a system that some researchers say has become increasingly complex and problematic. That's because the Internet of today is a very different place, having turned into a constellation of streaming and on-demand video content, e-commerce transactions, digital and social media, smartphone apps, and cloud software.
"You deliver a packet to the destination IP address specified in the packet. But you serve applications, which operate in terms of names," says Lixia Zhang, a professor of computer science department at the University of California, Los Angeles (UCLA). Zhang is also lead principal investigator of the Named Data Networking (NDN) project, heading up a team of researchers who have been working on a new Internet architecture intended to replace TCP/IP. Their goal is to create protocols that would support a single namespace and eliminate numeric IP addresses, enabling consumers and applications to access that data using only names.
It's a bold move, but it could significantly improve the way IP traffic is handled. For carriers, NDN could add resiliency to the Internet through better security and smarter routing and packet forwarding, according to Ronald Gruia, director of emerging telecoms research at Frost & Sullivan. For enterprises, it could mean the ability to move large chunks of data without necessarily identifying the address but simply pointing toward a certain topic, which would speed up information flow, he adds.
Labs are a world away from production environments, however, and NDN is still very much in the lab. But can they really do it? Could NDN wipe out TCP/IP, the foundation of the Internet? And more importantly, would the IT industry ever take it seriously?
From universities to ecosystems
Launched in 2010 as a Future Internet Architecture research project funded by the U.S. National Science Foundation (NSF), the Named Data Networking project was originally led entirely by computer scientists from a 10 institutions across the United States. It grew out of an earlier project, Content-Data Networking, architected by Van Jacobson, one of the primary contributors to TCP/IP.
David OranCisco fellow
The project has received roughly $15 million in NSF funding through 2016. But last fall, in response to growing interest in NDN from other research institutions around the world and the industry at large, the project expanded into the Named Data Networking Consortium. The project has opened its doors to several academic researchers outside of the United States, in addition to organizations such as the MITRE Corporation and several IT vendors, including Alcatel-Lucent, Cisco, Huawei, Panasonic, VeriSign and, as of December, Intel. For-profit companies must contribute $25,000 to join the consortium and obtain voting rights. The consortium, however, maintains that any protocols and architectures developed will be published as open standards.
"Cisco is good at managing technological transition points and maintaining its overall industry leadership in the networking space," Gruia says. "We see them doing that through participation in the NDN consortium. Obviously, their primary motive is to ride the next upgrade cycle when that happens."
The project's founders contend that as Internet traffic grew over the years, a single server and IP address could not fill the thousands to millions of information requests that people and applications made. The industry tried to resolve this challenge with anycast, which sends packets requesting data to one of many machines, and content delivery networks (CDNs), which replicate data across numerous locations that sit closer to the edge of the network. CDNs, however, still require DNS servers to look up and provide the requested information.
These approaches, while currently necessary to address the limitations of TCP/IP, make an Internet built on TCP/IP more complex as applications based on named data and content evolve, according to NDN proponents.
"The question becomes, is there a better protocol architecture that takes into account the changing nature of the applications that are on the Internet?" asks David Oran, a Cisco Fellow working on the project.
The NDN consortium is working hard to answer that question as it aims at building an architecture that could remove the need for DNS and would transfer content based on its name.
"It's an entirely new Layer 3 for the Internet, and it would replace TCP as well," Oran says.
Bringing multi-path routing to life
In an NDN-based network, data can have multiple locations natively. NDN protocols would decide what location to use to satisfy a data request through a combination of NDN routing and a mechanism called forwarding strategy. NDN forwarding is designed to enable multi-path routing, meaning that in an NDN router, a routing table entry could specify more than one output interface. Forwarding strategy decides which output interface to use.
"The default forwarding strategy that is most common in the NDN testbed today is one that keeps track of the performance of each interface, and [selects] the best performer while occasionally cycling through the other alternatives," explains Patrick Crowley, an NDN researcher and associate professor of computer science and engineering at Washington University in St. Louis, Mo.
Ivo Vachkov,IT consultant
When forwarding an NDN request for data, a router considers whether its packet buffers already contain the requested data, whether it recently forwarded a request for this data, and what output interface it should use to forward this request, says Crowley. If the router already has the data in a content store, which is an in-network cache, the NDN router can send the data immediately. If the requested data is already on its way, which the NDN router can find out using a Pending Interest Table, the router can aggregate further requests for the same data and reduce upstream traffic.
"This is a very efficient and natural mechanism for broadcast and multicast," says Crowley. The NDN router can determine where to forward the requested data based on a routing table that is very similar to an IP routing table, except it contains name prefixes rather than IP address prefixes, according to Crowley.
The request only needs to retrieve the desired data -- the actual source doesn't matter, as long as the packet is signed by the original producer, which "allows the consumer to verify that it gets exactly what it asked for," says UCLA's Zhang.
"[The request] does not necessarily reach the data producer -- data could be from a peer, a cache or the producer directly. The network performs the magic to bring the requested data back from nearest place," she explains. "NDN can do everything that IP does, and can do many more things that IP cannot do without additional tweaks, such as multicast delivery or in-network caching. NDN has them built in, but it takes IP multiple gymnastic jumps to support them."
Designed for video and security
Routers in an NDN-based network would have a "reasonably large" storage capacity to support caching. They would perform data caching that is automatically optimized in the middle of the network as needed, says Cisco's Oran.
Two motivations exist for this function. One is that movies and video comprise a growing portion of Internet content and are massive in terms of data. Similar to the principles of a CDN, an NDN-based network could cache that data closer to users to decrease the load on centralized servers and reduce latency. But there's another purpose for caching in NDN.
"In sensor networks that are power- and battery-limited, you don't want to ‘wake' the sensors all the time. You want the sensors to cache their readings in the network so that when someone needs the reading, you don't have to go all the way back to the device and wake it up again, using up battery life," says Oran. The ability to cache that data in the network, rather than on the device, would benefit the Internet of Things, which will include such sensor networks.
NDN proponents say their architecture would also secure traffic better than IP does, given how simple it is to spoof IP addresses. With NDN, when a device receives a reply to a request for named data, the data in the packet is structured to include a cryptographic signature from the publisher of the data.
"This enables the receiver to independently verify the binding between the name requested and the data received," says Crowley.
Can NDN really replace TCP/IP?
Outside of academic circles, however, some network engineers aren't so confident that NDN can usurp TCP/IP.
"In my opinion, it will take a lot of time for another protocol stack to reach the level of sophistication, adoption and quality of implementation that TCP/IP has," says Ivo Vachkov, a software engineer with a background in networking, who founded Xi Group Ltd., an IT consultancy in Bulgaria.
"I don't believe a new protocol stack will emerge to take the world by storm," he adds. "The amount of effort and money already invested will not allow for a new technology to be quickly adopted. Underlying technology stacks, data-link layer technologies -- Ethernet, mobile and DSL -- are even harder to change. It will be a while until those will allow us to migrate off of TCP/IP."
Members of the consortium acknowledge NDN is still in the early stages of research, and it's way too soon to say whether it will replace TCP/IP. If NDN does replace IP, it will occur very slowly, because a rip-and-replace job for something as big as the Internet simply can't be done.
"The deployment model for new architectures is to start as an overlay network on top of IP and then, over time, as the traffic shifts and the economics drive the change, you drop IP out and run [NDN] natively," says Cisco's Oran. "Most of these big swings take 20 to 30 years. We will still see IP as one of the main protocols for the Internet 20 years from now, but it may be sort of pushed down as a kind of transport for the new NDN architecture, if and when that new architecture gains traction."
As a new Layer 3 protocol, NDN would make packet-based requests for data similar to how the HTTP protocol makes requests for URLs.
"HTTP has proved to be very efficient, but an enormous amount of wasteful infrastructure is devoted to mapping HTTP traffic down to point-to-point IP channels," says Washington University's Crowley. NDN could eventually remove the need for that wasteful infrastructure and additional mapping, speeding fulfillment of data requests.
Yet the existence -- let alone the completion -- of a new Internet protocol would not ensure its success. Other alternatives that already exist still have not caught on, points out Vachkov, who is also a lecturer in networking at New Bulgarian University in Sophia, Bulgaria.
"Even attempts from a company like Google to promote a new Internet protocol, SPDY, which sits on top of TCP/IP and tries to address issues by using HTTP, are unsuccessful," he says.
Tutorial: Understanding TCP/IP
What's the relationship between TCP/IP and HTTP?
When DNS responses are slow, it means everything is slow