A cat working a bow tie

Breaking the Internet: how to hold it together

Is the Internet broken? It’s been pushed way past its design objectives. But amid incessant criminal activity and a growing data deluge, sticking plasters may still be enough to hold it all together.

If the Internet could sue for harassment, socialite celebrity Kim Kardashian would be first on the list to receive a summons. In March she once again threatened to break the Internet with a selfie. And once again, other than generating a wall of text from commentators, a staged photograph designed specifically to go viral barely resulted in a scratch to the core network. She should try long-form videos instead.

What the Kult of Kardashian cannot break, Netflix and its competitors might. By December last year, video and audio streaming accounted for more than 70 per cent of the total bandwidth used for delivering peak-time data to consumers across the US portion of the Internet, according to broadband networking specialist Sandvine. Netflix videos consumed more than a third of total downstream bandwidth, YouTube used 15 per cent and BitTorrent soaked up a quarter of the bandwidth going upstream.

Billions of dollars are spent funnelling copies of videos across continents and between them because people need to access them from particular locations. This is the argument that drives engineers working on a fundamental redesign of the protocols used on the Internet.

IBM CTO Bernie Meyerson says: “If you have a huge global network and you are streaming entertainment, you need to stage the content locally to the user or you are constantly having to ship this across the network.”

The original architects were not wrong in their choices – they were appropriate for a network where the communication is mostly peer to peer. Yet in an environment where most data fans out from one source into a multitude of streams, it may make sense to have a very different kind of Internet.

PARC researcher Van Jacobsen came up with the idea of Named-Data Networking (NDN) more than 10 years ago. It is a revolutionary proposal. He aims not to fix IP but to replace it. He argues the IP protocol set is the wrong technology for a growing number of applications that call on the ability to move data at high speed: “Our tools are working against us instead of for us.”

He proposed a system where users looking to view a video or grab a document use software that sends out ‘interest packets’ specifying what they want. Routers would, instead of directing the packets to known servers, act like the scattered pieces of a giant search engine. They would pass the interests into the network until a router says “a computer attached to me has this, I will pass it back down”. Routers might even have it stored in their own cache.

The beauty of a system like NDN is that popular content flows naturally out to the edge of the network. A router sitting in one of BT’s exchanges might well have a copy of the latest Justin Bieber video it can send straight to the user without digging deeper into the Internet. For the core network operators, less of their bandwidth is clogged up delivering the same data over and over.

Luca Muscariello, senior researcher at Orange Labs, says there is continued pressure on mobile operators to prove cost-effective services, which has caused them to look at information-centric networking (ICN) techniques such as NDN for more than just video. “ICN gives you a lot of possibilities for services and new service deployment. You can provide them at the edge of the network in a cost effective way.”

The downside to NGN is that it involves a root-and-branch reworking of the Internet. Service providers will have to install a completely new router infrastructure and it is unclear how NDN will deal with security.

Is NDN needed to deal with the data deluge? Muscariello says NDN is not the only option: other developments in networking may provide some of the answer. The move towards software-defined networking will help, he claims. The approach strips routers down to the bare minimum needed to forward packets to the right output port quickly. The intelligence moves into data-centre servers that react to changes in network behaviour and congestion, frequently updating the low-level rules the routers follow.

The pressure caused by services such as Netflix on core Internet bandwidth has already created a workable version of ICN based on traditional IP packets. Many high-traffic websites already rely on content-delivery networks of servers spread around the world that take on the job of delivering large files on behalf of the main servers.

A research project by a team of researchers from the universities Carnegie Mellon (CMU), Duke, Toronto and UC Berkeley together with Akamai and VMWare found that NDN works well for heavily requested content, but that the traditional model suits much of the ‘long tail’ – the large proportion of Internet traffic that is requested only once in a while. Deploying caches around the edge of the network – in effect a content-delivery network on a grander scale – would cater for most of the demand without the upheaval caused by a complete change of protocols.

Once a network protocol has momentum, it is hard to shift. Even though the future of TCP/IP was far from assured in the 1980s before the Internet became the core network it is today, the theoretically superior OSI protocol stack failed to supersede it. This was despite OSI having the backing of governments and computer giant IBM when it was at the height of its success. Even though up to that point “nobody ever got fired for buying IBM”, the company’s OSI-based Systems Application Architecture (SAA) won few converts after launch towards the end of the 1980s. The company was soon fighting for its own survival.

Even IPv6 has found it hard to make inroads on its near universal predecessor IPv4. It will take massive discontent with IP to shift it now. Even the inherent security problems that many of the protocols built on top of IP have may not be incentive enough as security on NDN is still a work in progress.

Recent articles

Info Message

Our sites use cookies to support some functionality, and to collect anonymous user data.

Learn more about IET cookies and how to control them