Cablecaster, September 2000

Consumer Window

Is "Giga-Fibre"

part of broadband's future diet?; Start-up research is showing that fibre to the home, using Ethernet technology, may soon be affordable



By Christopher Weisdorf, Rogers @Home Users Association

A little known company in Spokane, Washington, World Wide Packets, is in the midst of beta testing a product they call a "subscriber distribution unit", or SDU.

The SDU's role is to provision extremely high bandwidth services via optical fibre to residences and small businesses using the ubiquitous Ethernet protocol. A fellow member of the RHUA pointed out an article in the August 25th issue of PCWorld.com outlining the company.

According to the story, each SDU will be shared among eight individual residences or offices, provisioning regular Ethernet (10 megabits per second) to Fast Ethernet (100 megabits per second) per subscriber port. This is not unlike the current coaxial infrastructure, although the SDU model would appear to be far faster in theory.

The article goes on to say that each SDU will cost roughly US$1,500 and that individual subscribers will pay as little as US$40 monthly for this future broadband service. That would certainly put it in the same retail domain as cable and DSL service.

Now, other than the obvious benefits of "more for less" for customers and providers alike, what is particularly special about this technology? To me, it's the fact that some form of the Ethernet protocol is being used in every step of the SDU provisioning process.

Readers may recall how this space has extolled the virtues of Ethernet. I have not changed my mind and there are additional reasons why this datalink layer protocol is strongly favored above all others.

Right off the bat, most Ethernet proponents will state the protocol's primary assets to be ubiquity, scalability, compatibility, versatility, simplicity and cost. It's ubiquitous since the protocol is present in approximately 90% of the world's networks. It's scalable because all versions of Ethernet share the exact same frame types, along with very similar MAC (media access control) and PHY (physical) layers.

The Ethernets are compatible with each other for all the same reasons. It's versatile because it can support any application out there, including voice, video and data, and contains additional extensions to enhance the support of such applications. Finally, Ethernet has been adopted much more quickly than everything else because it's simple to understand and deploy, besides being very cost efficient. There are no other networking products on the planet that can deliver Ethernet's performance for anywhere near its (constantly dwindling) price.

Ethernet does, however, exhibit a number of imperfections. Detractors will often point out that when shared among too many users, Ethernet provides very poor overall packet latency and delay variation, as well as inconsistent upstream and downstream bandwidth. My own response to the above assertion is that any medium that is oversubscribed will suffer a similar fate. Due diligence must be taken to ensure that any shared medium which is Ethernet-based is not oversubscribed at any point.

A problem that is endemic to both Gigabit Ethernet and its 10-Gigabit Ethernet offspring is that of traffic management and auditing: It simply can't be done in real-time because the technology is just too darned fast. To put it into perspective, a Gigabit Ethernet link would take just over four seconds to saturate 512 megabytes of RAM, which is the maximum amount of memory that is supported on most current computers. When the performance of a network is supposed to be charted over the course of several weeks or months, this deficiency creates some definite problems.

Two other apparent downfalls of Ethernet is really only just one, with two implications. It's that it is a "best effort" protocol and, as such, contains no inherent facilities for QoS (quality of service), or virtual circuits. Ethernet has always taken an egalitarian approach to traffic, in that no one station may get priority over another, regardless of the type of applications in use, or how much they are being used. Protocols containing QoS features allow certain, "higher-tiered" individuals to gain network priority over their "lower-tiered" counterparts during periods of congestion.

My own opinion of QoS is that, in theory, if it is handled correctly it can be a boon both to service providers and customers. Unfortunately in reality, the overall subscriber base of a particular service has a much greater chance of suffering from very poor network performance due to QoS policy misappropriation and poor management of the underlying service by the provider.

An interesting fact to note about the above ramblings on QoS is that Ethernet does support an extension which allows the provider to establish various service tiers on a per-packet basis. Called "VLAN encapsulation" it's one of the methods that can be used to provision third party open access to broadband infrastructure.

The competing technology of ATM (asynchronous transfer mode), in contrast to Ethernet, does inherently support QoS and virtual circuit features, but not anywhere near the levels of bandwidth or the low cost and complexity associated with Ethernet. Indeed, some DSL vendors are moving away from the telco-favored ATM technology to Ethernet in hopes of speeding deployment.

It would be appealing to see a next generation of Terayon's TeraPro cable modem abandon the current ATM technology upon which it is based in favor of Ethernet. Although unlikely, such a move could save MSOs some money due to reduced costs in training personnel to provision and administrate the architecture. Deployment might also quicken for the same reasons.

The aforementioned corporation, World Wide Packets, is no stranger to Ethernet. Its chief executive was a co-founder of Grand Junction Networks which pioneered Fast Ethernet, and has had a great amount of influence over the networking industry and contributed greatly to innovation in the same sphere. Many others in the field will share the view that Cisco is where it is right now partly due to its acquisition of Grand Junction in 1995 (and for a now paltry sum of US$350 million, I might add).

World Wide Packets is also one of the founding members of the 10-Gigabit Ethernet Alliance, which contains names like Cisco, 3Com, Nortel Networks, Intel, Extreme and Sun.

While it is true that we are perhaps five, 10, or even more years away from a fibre-to-the-home scenario, that time will come. It's this type of technology, based on a concoction of the various Ethernets and optical fibre types, that may reign supreme as the communication infrastructure of choice in the future.

Although MSOs have a current edge over telcos for provisioning this type of technology the soonest, it will take someone with true leadership and vision to really start this ball rolling.

Christopher Weisdorf is the president and technical director of the RHUA.

Back to Contents




Copyright Notice Copyright 2002 Business Information Group. All rights reserved.