Network Computingwww.networkcomputing.com

Application-to-application flows are long-lived by their very nature. FTP transfers and HTML traffic are good examples. When an Ipsilon ATM switch sees FTP packets, the switch knows it may as well set up a virtual circuit because the transfer most likely will consist of enough traffic to make the VC worthwhile.

Client/server traffic is typical of station-to-station traffic. Once a client begins the login process, the client probably will have exchanged enough packets with the server for the Ipsilon switch fabric to recognize the flow and set up a VC for it.

Ipsilon tells us it has implemented both of these flo w types successfully, even in large ISP infrastructures. The company also says its technology has handled all traffic environments with ease. The third flow type, which in theory is similar to Cisco's tag switching and Cascade's IP Navigator, is designed for large ISP or carrier-class networks, according to Ipsilon.

To the rest of the network, an Ipsilon switch fabric looks like a big router, observing all usual routing protocols, such as the Routing Information Protocol (RIP), Open Shortest Path First (OSPF) and even the Interior Gateway Routing Protocol (IGRP). Complaints persist, however, that Ipsilon's is not the same kind of router that, say, a Cisco or Bay Networks router would be. For instance, it lacks much of the protection and other features that make Cisco's and Bay's products so valuable.

Ipsilon is addressing this concern by building Check Point Software Technologies' firewall software directly into its IP processing card in its gateway products. Check Point's firewall software will go a long way toward quieting those who complain about the completeness of the Ipsilon solution as a router.

The other big knock against Ipsilon is that its solution runs only on ATM switches and supports only IP. Ipsilon responds that it demonstrated IPX over its hardware at Comnet in February and that nothing precludes the use of frame-based switches. Ipsilon chose ATM because it was the only way to play in a multi-Gbps environment, which is where the company sees its products playing; frame switches are just too slow. Nonetheless, Ipsilon's current product offerings remain sparse, considering the level at which they seek to play.

For other legacy protocols, Ipsilon's solution is to leave the current infrastructure in place. That certainly works, but it may become a management headache, particularly if exi sting hardware is old and needs to be replaced.

3Com Corp.
3Com's entrant in the IP switching sweepstakes is FastIP. In short, if you are willing to accept a network interf ace (NIC) driver change on your clients and servers, 3Com has a good way for you to boost your network's IP throughput. This solution is intended for deployment in workgroup- to campus-sized environments; 1,000 or fewer nodes per FastIP network is a good upper limit. Rather than attempt to compete in the wide area, 3Com has chosen to partner with Cascade and IBM. Both are more than adequate bedfellows in the wide area.

The dominant NIC vendor, 3Com is leveraging its strength in that market to produce a solution that provides stem-to-stern performance improvement, as well as the possible (but as-yet-undelivered) benefit of application-level control over quality of service (QoS) requests. The advantage of 3Com's approach is that it leverages a lot of what you have in your network. That's good, b ecause some of this technology (specifically routing) is outside of 3Com's traditional realm of expertise.

Existing routers and switches can function in a 3Com FastIP environment; in fact, some router-- any router--is required to make it work. Intersubnet traffic must travel first through an existing classical, routed network. Once it has done that, a FastIP-aware NIC driver can set up a path through a switched network to avoid sending any more traffic through a router.

The switched network and routed network can and must live side by side. Whether that's a good thing or a bad thing depends on your network and how you want to evolve it. The obvious question here is: What do you do if 3Com doesn't have a FastIP driver for your client or server? Worse, what if you've got a Unix server and, hence, no way to put a 3Com NIC into it?