Is it time to move to 10 Gigabit Ethernet?

Large 24x7 organisations will find benefits, but most should wait.

Large 24x7 organisations will find benefits, but most should wait

For most organisations, 10 Gigabit Ethernet is still embryonic and too expensive to consider. The lack of several key support elements, including network management tools and TCP off-loading algorithms, is another reason not to be an early adopter. Research suggests price erosion will result in a 50% decrease in prices by 2005, with key support hardware and software in place at that time.

Otherwise known as the IEEE 802.3ae specification, there are several different options for this architecture since it supports three unique light frequencies as well as two different types of fibre interfaces. Some interfaces have been developed for wide area networks while others are more appropriate for local area networks.

Despite the high cost, they can be cost-effective in certain Wan configurations. One example is where a company needs 10x1gbps Ethernet links to connect sites 10 miles apart. Given the cost of leased fibre optic cables, there could be a tenfold saving in line costs, which far exceeds the cost of the 10 Gigabit Ethernet ports.

While few desktop applications require gigabit bandwidth, Intel has embarked on a plan to bring down the price of 10/100/1,000 adapter cards so that they match the current cost of 10/100 cards.

Assuming it achieves this price point, most firms will start buying the new cards, since it gives them room to grow at no additional cost. As companies add new desktops with faster buses and begin turning on gigabit to the desktop, there will be a need to aggregate these data flows, and 10 Gigabit Ethernet is the most likely choice.

The argument for using 10 Gigabit Ethernet in the datacentre is that is reduces the number of switches and layers required. It also uses less rack space and consumes less power. A datacentre with hundreds of servers each generating 100mbps of traffic has dozens of 10/100/1,000 Ethernet switches in a matrix architecture. In such cases the introduction of two 10 Gigabit Ethernet switches in active/active mode to replace the gigabit switches offers a significant decrease in complexity.

10 Gigabit Ethernet technology offers the speed and distance (up to 40km) to make IP-based storage area networks realistic - both for network-attached storage appliances and for iSCSI implementations. Ten gigabit links also provide adequate separation between a primary datacentre and a remote site for business continuity support as part of a disaster recovery strategy.

Initially, 10 Gigabit Ethernet performance will not be as efficient as 1gbps Ethernet because there are some missing technology pieces. Engineers are working on TCP/IP off-load, which means implementing part of an IP stack as hardware instead of software. The result would be faster performance.

They also are working on new algorithms for 10gbps packet inspection, but these are unlikely to be widely deployed until at least 2004. Suppliers also have no plans to offer protocol analysers for 10 Gigabit Ethernet, given the amount of memory they would require - they recommend that users troubleshoot slower connections at the edge of a network.

Another missing ingredient in 10 Gigabit Ethernet is traffic-shaping tools and another issue for early adopters is the lack of support from security technologies such as firewalls and virtual private network appliances. Also, some switching suppliers admit that their current chassis-based cards will only connect to their backplanes at 8gbps.

I agree with the assessment by most switching suppliers that 10 Gigabit Ethernet will follow the same pattern of adoption as Gigabit Ethernet: prices will remain high until there is sufficient demand to bring economies of scale - and that is not likely to happen before 2005.

By that time, Gigabit Ethernet will be pervasive to the desktop and 10 Gigabit Ethernet will be required for backbones. The earliest adopters of this technology have been forced to make that decision since their extensive use of Gigabit Ethernet switches left them no real alternative for aggregating traffic on an enterprise-wide basis.

Some enterprises with 24x7 operations will become early adopters because they see 10 Gigabit Ethernet as a way of providing continuing back-up of real-time transactions, and they will justify the cost by pointing to the unacceptable cost associated with downtime.

However, most businesses would be best served by ignoring the supplier hype associated with 10 Gigabit Ethernet and waiting until 2005 before any widespread deployments.

Stan Schatt is an analyst at Forrester Research

Read more on IT risk management