This question has to do with the confusion many readers express in trying to choose a LAN backbone strategy. There's Gigabit Ethernet, ATM, Fast Ethernet, IP switching - in short, no shortage of technologies, but no clear directions. Cut through the hype about all these offerings and lay out the real issues network managers should be considering in planning for the next-generation network.

Nolle: The biggest issue is 'incrementability'. Few companies are going to architect their LANs based on central planning decisions, whatever network managers might like to think. Projects funded by business units build networks, and that means pieces of the network move forward as local project activity justifies the move.

The best backbone technology will be one that can be added to existing networks without creating a lot of ripple impact and ripple cost. That means that ATM or Gigabit Ethernet can work, as long as both are about the same cost and as long as ATM's benefits don't require a major network shift in the ATM direction. Bits for bucks is the game, now and forever.

Gibbs: The real issue: Can you afford it now? To go for a major implementation now is most definitely to be donning the leather helmet and goggles of the test pilot. In six months to a year, the key issues of this new era of high speed backbone choices will start to become clearer: Which standards matter and which don't; which vendors will be stable (these two in combination are probably the biggest put-off to early adoption); and how problematic the technologies will be from the viewpoints of implementation and on-going management. That said, if you are in a large organization you should be trying this stuff out - looking for the opportunity gains that you live with and afford.

Heckart: There are really only three real issues that need to be considered in this type of purchase: cost, performance, longevity. What is tricky is that we tend to think about these issues as having absolute values and they don't - everything is relative to the customer's environment, goals, applications, budget, etc. What's 'good enough' for one company, or even one set of users, doesn't hold water for another.

The trick is to figure out what 'good enough' means and implement the solution that is cheap enough, performs well enough, and lasts long enough to meet your current and foreseeable needs. The problem faced by many users is that they try to figure out what's 'best'. Best changes every week based on their changing needs and the latest and greatest technology. Best never exists and certainly can't be implemented because by the time it is implemented, it isn't the best any more.

Briere: You know, too many managers are purists and continue to look for homogeneous answers when a mix is going to be the result. In any company, you might find ATM, Fast Ethernet, and Ethernet, or some other combination, because different sites and groups are going to have different needs. You don't buy Pentium class PCs for everyone in the company because everyone doesnt need one. You also don't need SONET to the desktop.

The most important thing to do is let the requirements drive the solution and not get hung up in the latest and greatest technologies.

Kearns: The majority of network connections are Ethernet and will continue to be. There s no gripping reason to go to a different technology for the backbone at this time. Ten megabits to the desktop and 100M bit/sec on the backbone works, and works well for many, many installations. Planning to add Gigabit Ethernet as the backbone with 100M bit/sec moving down to major trunks and, eventually, to the desktop seems a reasonable approach.

The 'gotcha' here is that for many networks, bandwidth isn't the bottleneck. Server/router/switch throughput, disk channel, bus speeds, local node buffering, and probably five or six other things all need to be considered. A pipe that s too fat simply wastes resources.

Bradner: I'd say the biggest challenge to network designers is the partial knowledge and strong convictions of their management. All too many decisions on the future directions of corporate networks have been based on broad generalities rather than an analysis of the actual needs of the existing network community. Someone from management reads the pronouncements of one of the big consulting companies - 'the answer is ATM (what was your question?)' - and decides on the future direction.

The real need is to do a technical analysis of the specific networking requirements and design based on the results of the analysis. Many technologies can prosper because not all networks are the same.

Go to the next topic...