Network Computingwww.networkcomputing.com

Cisco describes tag switching as a protocol-independent technology approach for large ISPs and carriers. Its primary benefits are route aggregation, improved performance and simplification of the view a router has of a complex network. Today, all routers must know about and talk to all other routers on a network. Tag-switching networks allow for the building of ATM backbones to which routers c an be attached, without the routers' needing to know the end-route details.

How It Works To achieve this simplification, routes are aggregated to tags with varying granularity. For example, a single tag might represent one application-to-application flow or hundreds of routes, effectively reducing the complexity within the tag switching environment.

In a sense, it seems to combine the Cascade and Ipsilon approaches, and the effect is similar. However, the approach is somewhat more general in that Cisco claims tags can be applied to many protocols. The N-squared problem is reduced to N with tags, and growth in the Internet requires only linear growth in the capability of the switched network.

Cisco's approach is aimed primarily at ATM or frame relay enviro nments where virtual circuits are available. There is no reason it could not be applied to packet-based networks as well, as long as some sort of virtual channel metaphor is applied on top of the packet-switched environment like 802.1p/q (see "What's in a Name,").

For campus-based networks, Cisco has taken its NetFlow technology out of the router and implemented it in the Catalyst 5000 switch. NetFlow is a proprietary means of performing Layer 3 switching, which Cisco brought to the Catalyst 5000 with the introduction of its new switching engine and 5500 chassis. Like other vendor approaches (including Ipsilon's approach), flows are set up only when they are detected. Once detected, however, the router is taken out of the equation and Layer 3 switching is substituted.

Ipsilon Networks
Ipsilon, as much as anyone, is responsible for the flurry of attention on IP switching. Interesting technology, combined with powerful and irreverent marketing, has helped Ipsilon g rab the notice of competition and media alike. Although Ipsilon's technology began as an enterprise backbone solution, the company sees no reason why it can't bring its technology into the wide area and beyond.

Ipsilon 's premise is that ATM and the standards that surround it are too complicated, have taken too long to develop and address many of the wrong problems. From throwing away the ATM Forum's standards to referring to the IETF as a debating society, Ipsilon steadfastly sought to go its own way. Eschewing the ATM Forum, where ATM is first and classical protocols are a distant second, Ipsilon takes the point of view that IP is king and ATM is merely a fast transport mechanism. That's a fairly bold posture. Ipsilon's products are coming to market and being field-tested now, so we can put the company's rhetoric to the test.

Although Ipsilon has put its specifications out in the public domain, the company is not about to debate the appropriateness of these specifications. Ipsilon is implementing prod ucts based on its standards and is happy to sign up partners willing to adopt these standards. Key among these standards are IFMP and the General Switch Management Protocol (GSMP), both available freely. GSMP is simple enough that Ipsilon claims it can be implemented by any major vendor in the span of less than a week. IFMP is designed to be equally as simple to implement.

How It Works The concept that drives Ipsilon's IP-centric thinking is that traffic should be divided into two classes: long-lived conversations (10 packets or more, for example) and short-lived conversations. Ipsilon has simplified the process and arrived at the same endpoint the MPOA committee will. Ipsilon's protocols set up virtual circuits for long-lived conversations only and use default paths for short-lived conversations. This is the essence of flow technology.