An alternative approach for net neutrality is proposed

FOR ALL ITS NOBLE AIRS, the FCC's draft of its proposed US net neutrality rules threatens to shatter the principle that the Internet is egalitarian.

Though all data is equal, the draft proposal says, some is more equal than the rest. Some data simply must be transferred more quickly and reliably than the Internet can manage in its present configuration, it argues. Television programmes must be streamed smoothly to be watched in comfort. Remote medical operations must be conducted without the least flicker of interruption.

If this new policy is adopted, it will signal a major shift away from the guiding principles the prevailed during the Internet's construction and which still underpin its policies today. (Virtual private networks (VPNs) do exist, but they are alternatives to the public Internet that are exclusively deployed by corporations and government agencies. VPNs are expensive to set up and support, and are often chosen more for security reasons than for latency or bandwidth guarantees.)

As the Internet doesn't discriminate between protocols at present, the consequences of such a change in US policy could mean a wholesale revision of the way it operates. It will, according to some, require the wholesale rip and replace of the entire Internet infrastructure. (A similar but prospectively far less disruptive proposal would see the build-out of a second networking infrastructure, called Internet2, that would parallel but not replace the existing Internet.)

The clean slate camp, as they are known, includes John Grant, who in the early 1980s developed the operating system for the classic ZX Spectrum. Later he's been the networking researcher who, with his Cambridge-based firm Nine Tiles, spent the last 20 years standing against the strong tide of Internet evangelism washing from the US west coast. He might characterise his resistance to the development of the Internet as similar to the church's stance against the tide of new age religious ideas that came on the same current.

Grant's proposals for a radically different method of managing network traffic take us back to the earliest days of the Internet to try to foresee how the future Internet might look after network neutrality rules are implemented.

They represent a compromise between forces that only 15 years ago were pitted in an ideological and commercial struggle for control of the Internet infrastructure. At the time, that was often characterised as a battle between Old Europe - the telecoms industry - on the one hand and the New World - the data networking industry - on the other.

It was ultimately an ideological battle between two methods of transmitting data that was fought over the principle of discrimination between different sorts of data traffic. And it looked for a while like the matter had been settled. The data industry's packet switched Internet Protocol (IP) was inexpensive, that is, unmetered, and ideologically sound - anyone could rely on IP to transport their data over the Internet on an equal footing with everyone else's data. In contrast, the Asynchronous Transfer Protocol favoured by the telecoms industry was expensive because it was developed and sold by an uncompetitive industry made up of highly regulated public utilities, and it was ideologically suspect because it was designed to give priority to traffic in which that industry had a vested interest, that is, circuit switched voice traffic.

Different strokes

While the markets were mesmerised by IP, Grant stuck with Asynchronous Transfer Mode (ATM), along with the telcos who ran it along the backbones of the Internet. His customers in places like the broadcast industries, such as the BBC, could rely on it to prioritise their time-critical broadcast streams. Most of them eventually let the tide take them to IP. But the limitations of the bare non-discriminatory Internet protocol have since become apparent to a wider audience.

"IP only gives them a good enough service under fairly controlled conditions. Quite often it has to be a closed network. So they have two networks. One for the Internet stuff and one for the media stuff," he says.

Grant does not propose a return to ATM, however. He sits on standards committees of the International Telecommunications Union, the body that originally hammered out the original ATM protocol and is now among those specifying standards for a future Internet. But ATM had turned out to be a disappointment anyway, and not because it was poorly specified.

"We felt that because [ATM] was similar to what we had been doing, we felt we understood it more than most people. It turned out that we didn't appreciate how badly everyone did understand it. What got taken to market was not really what was needed," says Grant.

"It wasn't implemented the right way. But somehow it got implemented as a packet network with lots of queues of different kinds of packets, and all the complication that finished up in IP as well."

Grant says he found a way around this dilemma, that neither IP nor ATM were good enough after all, when trying to develop a system for sending ATM cells over an Ethernet physical interface. Early this year Nine Tiles secured patents for a solution, called Flexilink, that attempts to settle the differences between the IP and ATM camps by combining their strengths and eliminating their weaknesses.

Best of both worlds

Flexilink has taken from ATM the method by which it guarantees a prearranged quality of service (QoS) in the transfer of data to someone, say, doing a remote brain operation. It does this, like ATM, by a connection-oriented method of routing data across a network. That is, it establishes a route across the Internet through nodes that can guarantee the required QoS and then drops the connection when it's done.

Crucially, Flexilink differs from ATM by dropping the much ridiculed, fixed 53-byte size of its cells (the parcels in which data is transported across an ATM network). Though routers on a Flexilink network will reserve bandwidth for a connection that asks for one, the variable cell size allows it to redeploy reserved capacity that is not subsequently used and that, under ATM, would otherwise be hogged by fixed cells. That spare capacity, along with anything else on the network that isn't on a QoS promise to a connection, is given over to IP traffic. The connections are rerouted if there are problems, or for the most crucial data flows, backup routes can be reserved from the outset.

This contrasts with IP most strikingly in the corresponding speed and ease with which the router nodes in the network can forward their data parcels. Not knowing in advance where data will be routed, routers on an IP network have to examine each packet of data and work out where to send it, buffering them to accommodate the fluctuations of service inherent in a system that works routings out as it goes along.

IP routers are consequently four times the size, consume four times the power, and cost four times as much as the equivalent switches used to route traffic on prearranged paths, the 100x100 clean slate Internet project, a collaboration of US universities and the US telco AT&T referenced in the Flexilink proposal, said in 2003.

"If you look at IP as originally conceived," says Grant, "it is all delightfully simple, and it's simple because there are so many things it doesn't do". It was for the same reason supposed to be cheaper, with routing decisions done in software. Instead, they have to be hard-wired, he claims.

QoS, meanwhile, wouldn't be a problem because network capacity could be cheaply expanded to create enough room for fluctuating traffic levels. Yet QoS controls have had to be laid over IP to bring its 'best effort' guarantee up to levels demanded by time-sensitive applications. Grant calls them "sticking plasters" that merely increase the complexity, cost and power consumption of the network.

Future Internet

One alternative is for a clean slate approach like Flexilink to replace each and every node, adapter and software application in the entire Internet, which Grant says only sounds less feasible than it really is. Flexilink may not be inevitable, but it is certain that the future Internet will have some more certain paths laid across its apparently chaotic landscape. A route and branch upgrade will just take time, he claims, apparently confident in his faith that the world will eventually come around to adopt his ideas.

Grant sees Flexilink becoming established in "islands" operated by high-demand users. They would anyway not be able to apply its unique routing logic across parts of the Internet that were not lined with Flexilink routers. And that is how complex networks have to be changed. Grant happens to be chairman of the IEC62379 standards board, which is creating a management control protocol for professional audio equipment. He says the BBC is requesting that audio equipment manufacturers conform to the nascent standard, which requires the network to give it a quality of service commitment. That might provide some leverage.

Hence islands may form, if Grant and his like-minded colleagues at the ITU exert their influence and are successful. As long as future Internet technologies like Flexilink accommodate IP as well, the result might be a viable compromise, structurally as well as idealogically, between the rigidly ordered, exclusive network the telecoms industry would have created had it always had its own way, and the adaptive reality of the Internet today that grew up based upon the DNA of IP.

If US regulators redefine network neutrality to include a multi-tiered overlay on the existing Internet, they will have to ensure that in allowing certain data traffic to demand priority, they don't so overwhelm weaker network resources that they seize up from congestion.

Whereas some European future Internet proposals do suggest that telcos will be required to provision enough capacity for IP traffic to thrive after they have allocated space reserved for 'priority' data traffic, the draft US proposals contain no such requirements.

So there is certainly more planning work to be done, and various factions will contend to determine what the future Internet will become. More battles await, and the stakes for the Internet could not be higher. µ