Supplier dilemma for Learning Grid

It was a crisis of identity. During the launch of Digital Network's face a variety of issues that will cause problems. One of the most crucial is whether the service provider should be national or regional. How are Britain's 32,000 schools going to be networked? Last week Network News shed some light on this matter after wading through the Government's consultation document, Connecting the Learning Society.

The Blair/Gates summit clarified the points that Internet connections would be made using BT ISDN2 lines and the Government will put up #100m to help fund the project. But, the issues of whether a single network service provider should be awarded the contract on a nationwide basis, or whether regional contracts should be issued still remains. As does the question of money.

Tony Blair's plans to connect Britain's schools to an Internet-based National Grid for Learning has sparked a debate within the network service provider community. Phil Hemmings, head of corporate affairs at Research Machines (RM), which is the main supplier of IT to the UK education sector admitted: "It will get very competitive as the IBMs and ICLs of the world see this as a money making opportunity. Schools will have no allegiance to a certain network supplier, be it 3Com or Cisco, but will go for the most cost-effective solution."

The public has already been told that the Grid will be a 'loosely linked mosaic of networks based on the Internet', which appears to be extremely vague. Hemmings coined the phrase 'blurry'.

Professor Stephen Heppel, director of Ultralab - an education project connecting seven schools in the South East - at Anglia Polytechnic University asked: "What is the National Grid, an ATM network?"

Heppel, who has headed Anglia's Ultralab educational technology research centre for five years, said: "The National Grid for Learning is a very strong term and inconsistencies are bound to happen. But there are many layers the Government has to consider.

"First some kind of pervasive network is crucial, then service provision must be considered. Content for the National Grid also has to be considered, and the tools necessary for contributing to it. There is no use being given content if you can't contribute to it," Heppel explained.

He also pointed out that: "The Government should give every child in the UK an email address or the children will turn into couch potatoes.

Also a regulatory framework will be needed."

The Government now faces a difficult decision. If it decides to opt for a nationwide franchising approach whereby all schools will be obliged to consider the same level of service at the same cost regardless of location, the consortia would just consist of the largest IT companies able to offer full UK coverage.

This, as the consultation paper points out, "could be a deterrent to potential new entrants to the supply market". A regional or local franchising approach might be more attractive, allowing smaller companies into the market and presenting less risk of an over dominant supplier.

However, difficulties may arise for schools in areas too isolated to be attractive to consortia. A third option would be a combination of nationwide, regional and local franchises.

A further difficulty is that schools have no definitive IT infrastructure.

RM provides 5,000 schools in the UK with internal Lans running on 10Mbps Ethernet, other schools have virtually nothing.

Bromford School in Wickford, Essex, is an exception. The school has spent #1m on its IT infrastructure by linking 200 PCs via the cable company Telewest to the Internet using a broadband access programme. John Poad, headmaster at the school, said: "Blair has pledged an extra #100m for schools but it's just not enough. Each school would get on average #3,000 which is peanuts in IT. The National Grid is a reasonable concept but more money is needed to make it effective."

Surprisingly, the Government's consultation paper agrees with Poad. It states that more money and funding is needed to provide schools with the necessary hardware and software, and to develop their systems once the Grid is operational. But no decisions have been reached.

The only concrete decision reached by the Government is that schools will link to the Grid using ISDN2 at 64Kbps. That leaves network providers and pupils waiting. GIGAswitch/Ethernet is not exactly its own, in fact a comparison reveals it was originally a Prominet switch. So what is the reason for the secrecy? new Giagabit Ethernet switch for backbone applications - the GIGAswitch/Ethernet - Nick Gosshawk, Network Product marketing manager at Digital, insisted: "It's a Digital switch. It is a real member of the GIGAswitch family."

Under cross examination, Gosshawk would neither confirm nor deny that the GIGAswitch/Ethernet switch was actually an OEMed Prominet Cajun P550 switch. He continued to stress to the gathered press that it was indeed a home-grown Digital switch. A simple swift comparison of specifications revealed Digital's little secret.

In Reading later that day, Neil Wilkins, Digital's UK marketing manager admitted it was an OEMed switch, but preferred to be silent about the source, except to remark that it was a joint development with an unnamed Gigabit Ethernet start-up company. Although "joint development" is stretching credibility to the limit for just carefully sticking a Digital label over the Prominet one.

The reason for Digital's reluctance to admit this is a badged product?

Senior analyst at Datapro, Marina Smith, has an answer: "If people know it is a Prominet switch, they would also be aware that they could probably buy it cheaper from Prominet if they do not mind foregoing the better support that Digital offers," she suggested. "Prominet has only recently set up offices in the UK after all."

Prominet was formed in 1995 by a group of ex-Chipcom executives following its acquisition by 3Com. Since then, Prominet has focused on switches that serve the campus backbone.

Its Cajun P550 switch (the Digital GIGAswitch/Ethernet switch) is claimed to be the largest Gigabit Ethernet switch available. And, with a backplane capacity of 41.6Gbps that provides a throughput of 20.8Gbps, it probably is.

It is certainly larger than most ATM Lan switches, with the exception of the Cabletron MMAC-Plus. The seven-slot switch can be configured with a mix of 10/100/1000Mbps Ethernet ports; and with 12 Gigabit Ethernet ports, can forward 18 million packets per seconds (pps).

The multilayer platform can also perform Layer 3 switching at 10 million pps. Prominet is attempting to provide Quality of Service (QoS) through support for both IEEE 802.1p priority signalling and 3Com's proprietary Priority Access Control Enabled (Pace) technology which provides priority queueing with flow control.

Gosshawk believes that this is an enormous breakthrough for Ethernet.

He said: "It can now be argued that many network managers can deploy Ethernet across their entire networks." Gosshawk continued: "Ethernet is now the dominant force in the market and for most applications, everything else is now a legacy or niche technology. And that includes ATM in the Lan."

But Datapro's Smith disagreed. "Unlike Nick Gosshawk, I don't think network managers can now plan their entire network around Ethernet," she said.

"There is a good case for campus backbones to be built around ATM." For many, ATM will still be the preferred choice for large networks (see box below).

Smith added: "ATM within Lans will only ever be adopted by those with a real need for both high bandwidth and QoS. The QoS available for Ethernet was only ever going to be as good as ATM has right now." There are still a lot of questions surrounding Gigabit Ethernet (see box left).

With Gigabit Ethernet, two basic vendor strategies are emerging. The first is the simplest, the most pragmatic and thus most likely to succeed.

This strategy uses Gigabit Ethernet for aggregating multiple 10/100 Mbps Ethernet channels on a single high-band-width link. Key elements here are NICs, repeaters and Layer 2 switches.

In this scenario, Gigabit Ethernet is deployed gradually at key points in the network, but does not seek to replace the corporate backbone.

The second strategy is much more ambitious. Any attempt to attack the backbone market has to address the router problem - current high-end routers cannot handle Gigabit interfaces. Companies like Digital and Prominet have developed Gigabit routing switches operating at both Layer 2 (switching) and Layer 3 (routing), which is where routing functions are performed.

From the customer's perspective, the problem is that all so-called Layer 3 switches are proprietary (vendor-specific) implementations. The only exception to this is ATM switches using the new MPOA specification.

For the network planner, an early adopter of multilayer Gigabit Ethernet switches runs the risk of vendor lock-in from a small number of start-up firms, any of which could be acquired tomorrow.

Perhaps Digital will acquire Prominet and make the GIGAswitch/Ethernet switch a real Digital switch, after all.

GIGABIT ETHERNET

Common criticisms

- Not the same as Ethernet. True - the frame format is the same, but the physical layer is Fibre Channel with changes to MAC and repeater specifications.

The Gigabit Ethernet camp may be fibbing when they claim it is the same.

For the customer this doesn't matter if the upgrade path is compatible with the installed base.

- Incompatible with existing equipment. Many, if not all installed Ethernet switches, do not support Gigabit Ethernet interfaces. In most cases, they will require a forklift upgrade. New management and tools will also be required.

- Does not support most cabling systems. It only runs over fibre. A study group for transmission over UTP with a goal of 100 meters has been formed.

Meanwhile, anticipate cabling system upgrades.

- QoS is difficult to implement. A massive software undertaking (RSVP+RTP+RTSP+801.1p+802.q+802.3x) will be required to mimic ATM QoS on a connectionless network. Most of the standards mentioned here are still not ratified.

- Lack of standards-based products. The first wave of GE products will be pre-standard, based on 802.3z draft specification. Many are implementing proprietary feature sets which will cause vendor lock-in and limit future interoperability.

In some cases, new silicon will be required to bring products into conformance with future specifications.

FACT FILE

Ethernet v ATM

For the desktop 10/100Mbps Ethernet is most common, and, in single buildings, 100Mbps Ethernet, while ATM is preferred for multi-building campus networks.

When trying to compare costs, network managers should compare the total solution or system cost, rather than just the technologies.

Gigabit Ethernet:

- One building

- 500 dedicated users on switched Ethernet

- 26 servers on Fast Ethernet

- Flat network, one subnet

- Cost: $300 (#187.5) per desktop

ATM:

- Larger building campus

- 1000-2000 users per building

- 250 servers on ATM OC-3 (155Mbps)

- Redundancy and loadsharing for switch connections

- 20 subnets

- Lane and one-armed routers; later, MPOA

- Cost: $300-$600 (#187.5-#375) per desktop

The conclusion is that for large networks, the equipment cost on a per-seat basis will be roughly the same for both Gigabit Ethernet and ATM.

This is because the price of large networks is dominated by the price of the edge switches, so the type of equipment in the core is less significant.

Based on experience, ATM is and will be the preferred choice for large networks.