Networking - King of the castle

In ATM'S early days many industry watchers forecast a great future for this cell-based technology. It was not just better than frame-based technologies like Ethernet at guaranteeing quality of service for simultaneous voice, data and video services, it was seen as the only technology that could achieve this.

As the world moved on to multimedia enabled applications, ATM's domain was expected to grow until it naturally extended from the enterprise backbone to the desktop.

Experts also pointed out that since the giant telecoms carriers were standardising on ATM for the WAN, deploying ATM within the enterprise would give user companies the benefits of a seamless ATM environment, with no troublesome protocol emulation anywhere. The technology argument was compelling.

All that remained was for the 'per port' price to lose its steep, early implementor weighting and become competitive with Ethernet. What could possibly go wrong?

Quite a few things, as it turned out. Several events disturbed this vision, and none of them had much to do with the virtues of ATM as a technology.

The scheduled autumn launch of the first wave of Gigabit Ethernet products is the latest symptom of a process that has made it deeply unfashionable to predict dominance for ATM.

The launch of Gigabit Ethernet, according to major players such as 3Com, Cisco and Bay Networks, will have two main outcomes. It will offer a plausible alternative backbone technology to ATM (FDDI is already sliding into third place in new backbone installations), and it will provide a very attractive server level distribution medium - again in competition with ATM.

Even before Gigabit Ethernet, two major developments in Ethernet networking had done a good deal to limit the growth of ATM to WANs and the campus backbone. These were the introduction of switching to fine-tune LAN bandwidth, and the launch of Fast Ethernet, which offered ten times the capacity of standard Ethernet (100Mbps as opposed to 10Mbps).

Since around 75% of the global installed base of PCs are on shared media right now, many companies could - and still can - solve any short-term bandwidth pressure problems by micro-segmenting the LAN. Installing a switched topology can create dedicated 10Mbps pipes for specific groups of users and for individual users. Fast Ethernet takes this whole process a step further by offering 100Mbps links either right to the desktop - or more commonly, from the server to the workgroup - with switching being used to create dedicated per-user 10Mbps pipes.

However, there aren't too many applications being used by either the private or the public sector right now that demand even 10Mbps pipes to the desktop. For example, MPEG II quality video, which is still awaiting the roll-out of sufficient numbers of DVD-capable PCs, requires only around 4Mbps.

Given the ready availability of a cheap standard Ethernet alternative, it is hardly surprising that enterprises aren't rushing to install ATM to the desktop.

This is also why 100Mbps Fast Ethernet has found favour largely as a good-sized pipe between servers and switches, rather than being run out to the desktop. There have been some 25Mbps ATM desktop implementations, largely by organisations with quality of service concerns for such things as projected desktop video conferencing solutions or computer-based training, but so far numbers have not been significant.

Here's the real question for Gigabit Ethernet products: will organisations which would otherwise have gone for ATM in the backbone now move to the latest version of a technology they know and understand, further inhibiting the roll-out of ATM? The leading vendors find this one too close to call, and will be betting on both technologies simultaneously.

As Cisco consultant Dave Ginsberg puts it: 'We'll offer both ATM and Gigabit capabilities and let the customer decide - there's no reason for us to take a religious view here.'

Cisco already has a bandwidth expanding alternative to Gigabit Ethernet in its Fast Ethernet Channel product, which aggregates four full duplex 100Mbps Fast Ethernet pipes to provide a total bandwidth capability of 800Mbps. In Ginsberg's view, Ethernet in its faster manifestations has no natural end term to its life. 'We're seeing a number of customers using Fast Ethernet in the distribution layer and ATM in the backbone. There's nothing stopping organisations from implementing mixtures of technologies for the foreseeable future,' he says.

Bull consultant Richard Holden is one of the few who see Gigabit Ethernet as a passing distraction. 'Our view is that going for Gigabit Ethernet will just postpone the inevitable. ATM is the way forward, and installing Gigabit Ethernet will simply provide users with a four to five year leeway before developments in applications force them to move to ATM anyway,' he claims.

Holden points out that while Gigabit Ethernet offers scalability with today's applications, its inherent limitations as a collision avoidance protocol creates unavoidable unpredictability under pressure. 'You know it will fail, you just don't know where and when,' he notes.

While confirming that Bay Networks will be launching a Gigabit Ethernet product at the end of the year, product manager Paul Trowbridge points out that in addition to quality of service issues, Gigabit Ethernet is hampered by Ethernet's spanning tree management protocol. The failover time in spanning tree topologies can be up to 90 seconds. This is the time it takes for the network devices to reconfigure themselves, identify where the break is and diagnose how to route around it, he explains. Many applications will time out during a 90 second delay. ATM, on the other hand, will deliver a reconfigured set-up in five seconds maximum.

Despite this drawback, Trowbridge expects steady growth in demand for Gigabit Ethernet as a pipe between servers. However, in his opinion its deployment will be held back by the many 1Mbps-4Mbps switches currently in service in enterprise networks: 'Since most switches have a maximum bandwidth of 4Mbps, this will be a bottleneck until the next generation of silicon switching feeds through - which won't be until well into 1998.'

Trowbridge expects the implementation curve for Gigabit Ethernet to follow the pattern set by Fast Ethernet. 'It began to be talked about in 1992, and the standard reached maturity during 1993 and early 1994. But it wasn't until 1995 that we saw real volume product at commercially viable prices coming to the market, while the boom took until last year to hit,' he says.

3Com UK product marketing manager Joe Frost acknowledges that Gigabit Ethernet's lack of prioritisation and class of service capabilities are drawbacks when it comes to the next generation of applications. However, he points out that the expected price differential between Gigabit Ethernet and 622Mbps ATM, plus Gigabit Ethernet's familiarity to many organisations, will encourage many to install it as a backbone technology instead of ATM.

'Today, many enterprises are using Fast Ethernet as a backbone,' he notes.

It is nowhere near as reliable as FDDI or ATM, but it still does the job for small to medium sized networks. 'We are looking to win the same kind of market share with Gigabit Ethernet as we currently have with Standard and Fast Ethernet,' Frost says.

However, he points out that demand for combined ATM/Ethernet networks is also likely to grow. 'The US market forecasting house, Tam Delloro, are predicting greater than 800% growth over the next 12 months for Ethernet to ATM switch shipments - and our strategy is also to follow that prediction,' he says.

Frost says that organisations thinking about installing Gigabit Ethernet will need to look at the current state of their network cabling infrastructure.

Gigabit Ethernet is not simply a case of more of the same.

There are four versions of the specification. Most people will be going for the 1000Base-T spec, where the goal is to run 100 metres on Cat. 5 cabling for workgroup implementations. Frost believes this will not be available until 1999. The main early use will be over fibre, where 1000Base-SX multimode will run to 500 metres, while 1000Base-LX will go out to two kilometres on single mode fibre.

'These features on fibre will dictate the use of Gigabit Ethernet, and the dominant use will be server-to-switch and switch-to-switch connections,' says Frost.

The two technologies will continue slugging it out. ATM has reliability on its side, but Gigabit Ethernet can call on the crowd's support. It remains to be seen which will win the title.