Here you go.
Why Stupid Networks like the Internet are a Smart Idea
April 8, 1999 - 7:45 PM
By Roderick Beck
Introduction
In June of 1997, David Isenberg, a former Bell Labs employee, created a furor at AT&T (T) when an article written by him was published without his permission on the Net. The article, entitled "Rise of the Stupid Network", argued that traditional phone networks, including AT&T's, were quickly becoming obsolete.
According to Isenberg, the traditional network was being superseded by Stupid Networks, the best known example being the Internet. In this article, I explore the controversy surround the idea of stupid networks.
Technology Primer
The Public Switched Telephone Network (PSTN), which handles virtually all phone calls, is based on the idea of circuit switching. This means each phone call involves creating a circuit between the parties. This temporary circuit is like a private road: it reserves transmission capacity for a single task such as carrying a conversation or a fax.
This is highly inefficient. A phone call typically uses only 30% of the circuit's capacity because of pauses between words and the fact that only one person (hopefully) is listening.
In contrast, the Internet and most data networks rely on packet switching, which forces individual circuits to multitask by simultaneously carrying all sorts of unrelated traffic such as a phone call, an email message, and Web pages. The result is a much more efficient network with circuits kept closer to capacity.
In order to share a circuit for different uses, signals are broken into small digital packages, called packets, whose granular nature helps to keep transmission pipes full. In addition, packet switching increases efficiency by distributing traffic among different routes.
For example, a Web page will be decomposed into packets and then sent by different routes to its final destination, where it is reassembled. This helps to equalize traffic across the network, lessening the likelihood of transmission and switching bottlenecks, and reducing the need for expensive capacity.
The Heart of the Matter
What distinguishes the PSTN and the Internet is that the former is very smart and the latter very dumb. Yet the irony is that smart networks are being eclipsed by their dumb relatives.
Circuit switches are highly intelligent specialized computers: they set up calls by talking to other switches, do database searches to route 800 calls, record information for billing purposes, provide services like caller ID and perform a multitude of other tasks. In essence, the circuit switches create the value in a network.
Not surprisingly, the switches are very expensive, costing as much as $5 million a piece, and are incredibly complicated. At the other extreme, packet switches are very simple, but very fast routing devices that simply 'ship the bits'. Consequently, their expense is a tiny fraction of the former. While it is clear that packet switching is more efficient than circuit switching, intelligence should be circuit switching's strong suite.
However, what David Isenberg realized in the course of working at AT&T was that too much intelligence was a bad thing. Isenberg was assigned to a project to improve the sound quality of AT&T's voice network. It became very clear that sound quality could be improved by boosting certain frequencies of the sound signals.
On a packet switching network this could be easily done by upgrading software, an easy and cheap solution that required no modification of the network's underlying hardware. For example, IDT has improved the sound quality of its Internet telephony service by modifying the software that users install on their PCs to make calls.
But in the case of AT&T's long distance network, Isenberg discovered that most solutions required the unacceptable alternative of redesigning much of the hardware. The crux of the problem was that network intelligence was hardwired and network components were tightly integrated. Isenberg was also struck by interesting disparities between dumb and smart networks.
Phone networks have dumb terminal devices - an ordinary phone simply converts electrical impulses into sound signals. In contrast, the Internet relies on increasing smart terminals like PCs and workstations that are driven by the dramatic growth in computing processing power. Intelligence in the PSTN lies in the center of the network, whereas it resides on the edges of the dumb networks. Furthermore, he noted that data networks were growing many times faster than the traditional networks with Internet traffic doubling every couple months.
In the end, the Net would be a giant, the PSTN, a dwarf.
Isenberg's conclusion was straightforward: the increasing sophistication of terminal devices meant that networks no longer had to be smart, they just had to be fast in order to carry the plethora of data that users want. In other words, intelligence was migrating from the heart of the network, the switches, to the periphery where the smart terminals like PCs, Webphones, and Palm Pilots were located.
The heart of the network was becoming a simple transport mechanism which was much more efficient than the traditional circuit switched technology and also much more flexible. The great flexibility of dumb networks came from their reliance on the Internet Protocol. This protocol, a sort of language for computers, allowed any packet switching technology to carry any type of information, regardless of nature or format.
Fundamentally, IP packet switching networks don't give a damn about what they are transporting, while intelligent networks are extremely fussy. This means that offering a new service is difficult for the PSTN because it may not be compatible with the current hardware. In contrast, the Internet protocol ensured that the packet switching would accept virtually anything given to it.
The other key advantage was that an IP packet switching could handle telephony, raising the possibility of enormously simplifying telecommunications. Instead of dozens of different voice and data networks tailed to the specialized needs of customers, it was now possible to envision an AllNet: a single integrated network to provide voice, data and multimedia.
The Holy Grail had been discovered.
Not surprisingly, Isenberg's message was not warmly received by AT&T's network management. It was a direct repudiation of their life's work: perfecting the PSTN. Although Isenberg kept his job, management rejected his desire to speak at public conferences and he eventually left to work as a consultant.
Interestingly enough, AT&T recently vindicated Isenberg's thesis that dumb networks were indeed better by announcing in early March that the company would no longer buy circuit switching equipment, but instead focus on building an IP data network.
However, it is still not completely clear how dumb the networks of the future will be. Most Internet phone calls do not use the Net. The reason for this is that the Net lacks the intelligence to provide the quality that is demanded by voice communication.
Instead, most Internet calls use privately managed IP networks that have more intelligence built into them. Another interesting fact is that AT&T's IP network uses a packet switching technology that builds 'virtual circuits' - in effect imitating the PSTN it will replace.
Isenberg's view is that this 'virtual PSTN' is a mistake. The key to guaranteeing quality is simply to throw capacity at the problem. When dumb networks become congested, everything, including time-sensitive traffic (like voice and streaming video) suffers. But capacity is becoming increasingly cheaper. Packet switches are computers, and Moore's Law says that their power doubles every 18 months.
A new technology called Wave Division Multiplexing can increase the carrying power of a fiber optic cable by 11,200 percent. Since congestion means insufficient capacity, Isenberg says one should over-provision the network.
Keep it dumb, but put it on steroids.
ragingbull.com |
