The following text is copyright 1997 by Network World, permission is hearby given for reproduction, as long as attribution is given and this notice is included.

Thinking outside of a box

ATM put lots of people in a box. Almost all of the pundits, big consulting firms and trade press writers spent the last few years saying that ATM was the inevitable future for all types of networks. Campus backbone networks, wide area networks and even LANs were going to be running over ATM any day now. This widespread assumption has meant that a realistic evaluation of what technology should be used in what situation has been very hard to do in many environments. Corporate IT managers read the stories and became believers that the best strategic direction to head in was towards a seamless ATM world. Any suggestion from the people actually running the network in favor of some other technology for some part of the network was seen as diverting the path towards the preordained future and could not be seriously considered.

For a number of reasons (many explored in previous editions of this column) ATM is no longer being seen as quite so inevitable a total solution. This has made it possible for some network designers to start actually trying to figure out what the best technical solution might be for particular network problems. This is not yet true everywhere but there are more people getting out of the confines of an assumed ATM future all the time. And there are some interesting technologies to look at. I'll use interconnecting network servers and storage systems as an example of a problem area.

Just because ATM may not be the answer to all problems does not mean that it might not be the answer to specific problems. High performance ATM interfaces are now available for many types of servers and storage systems. If LAN-like segmentation is desired (it is by me) ATM systems can run the ATM Forum's LAN emulation (LANE) or Multiprotocol over ATM (MPOA) or the IETF's Classical IP over ATM,

FIberchannel is a standards-based high speed networking technology for this type of application that can support long distance links and large packet sizes. (Large packets are more efficient for bulk data transfer than the smallish 1518 byte maximum size on Ethernet).

Gigabit Ethernet should also be considered. Gigabit Ethernet looks like it will be a quite cost effective solution in this area but support for large packet sizes, while present in some products, is a non standard feature.

I've seen systems designed to extend the disk-to-computer SCSI interface into switches and to use that as the basis of a network to interconnect a server farm.

Finally, the somewhat silly talk about fast token ring as a new basic networking advance would not have happened without the change in attitude about ATM.

I'm not trying to list all of the options or to make any specific recommendations but to point out that once you free yourself from an assumed answer you can actually figure out what technology and design will do the job best. This may be a new experience for a lot of network designers.

disclaimer: Far be it from me to dissuade those who assume that Harvard is the answer but the above observations are my own..