For the first time in a long time, network managers have a real choice of products from which to build their next-generation campus backbones.
But the choice between Gigabit Ethernet and ATM is not easy, and it is not clear whether frames or cells will prevail.
Gigabit Ethernet backbone switches with massive backplanes, high-port densities, low latency and relatively low prices are starting to roll off manufacturers' assembly lines and into customer sites. (Even though the IEEE standard is not officially in place.) These switches can be dropped right into existing Ethernet/Fast Ethernet networks and can offer network managers a no-pain backbone upgrade option. At the same time, however, we are experiencing something of an ATM renaissance. After literally years of painstaking standards definition followed by complex code development, powerful, sophisticated and relatively easy-to-use ATM systems are ready for deployment. And unlike their frame-based brethren, the devices are architected to handle constant bit rate (CBR) services such as voice and interactive video on Day One.
In my view, it is in this area of 'multi-services' that the frame vs cell battle will be fought and won. While ATM certainly can provide the backbone for a pure data campus network, much of ATM's power would remain untapped. For such traditional data networks, fast frames -- Gigabit Ethernet or High-Speed Token Ring -- will do the job.
For network managers intent on collapsing cross-campus or cross-country T-1 PBX links together, ATM clearly is the only viable option.
Streaming video -- such as playing back a training CD-ROM -- can be handled easily by cell- and frame-based networks. Surprisingly, bandwidth draw from such applications is quite low and latency is not usually an issue.
Interactive video is latency-sensitive and would, in theory at least, be better suited for ATM's CBR service than for transport over frames.
Many products demonstrate extremely low latency that, combined with raw bandwidth, should allow even CBR-type traffic to be handled satisfactorily across the switches. Still, interactive video remains a minor application for mostand is unlikely to sway the frame vs cell decision.
The wild card, in my view, is packetised voice. Referred to as voice over IP or voice over networks, the success or failure of this movement will significantly affect the frame vs cell balance of power.
While voice is the classic example of a delay-sensitive, CBR application, the dramatic economic advantages of shipping that traffic over usage-insensitive data networks has made this an extremely hot area.
Network managers would do well to watch the voice-over-IP market closelyfor indicators as to whether the future belongs to frames or cells.
Kevin Tolly is president of The Tolly Group, a strategic consulting and independent testing firm in Manasquan, New Jersey.