Network Computingwww.networkcomputing.com

Enterprise customers using ATM for their WAN connections have been at the mercy of carriers for years and have been relegated to using only PVC-based connections, as carriers have been slow to adopt any new transmission options. This has forced those customers to build networks that would switch packets and to pay for PVCs and bandwidth that may not be in use all the time. SVCs are meant to eliminate the need to pay for unused bandwidth, letting enterprise customers use the network only when needed. Creating an ATM network based on SVCs has been slow in coming to most customers except those who can afford to lay their own fiber or lease fiber from a carrier.

With ATM SVCs, an enterprise can route cells across the network, building the connections between switches as cells go. Customers using private ATM networks or SVCs on carriers' networks can choose IISP (Interim Interswitch Signaling Protocol) or PNNI (Private Network-to-Network Interface) to route data. Network flexibility can be expanded even more by joining the routing capabilities of IISP or PNNI with MPOA (Multiprotocol Over ATM) or MPLS (Multiprotocol Label Switching). With these, IP packets can be routed over a WAN, and customers pay only for bandwidth used. QoS is thus maintained.

A Working Stopgap

IISP was created in late 1994 as an extension to the UNI (User-to-Network Interface) 3.0/3.1 specifications enabling a simple network context. In simplest terms, IISP is to ATM as static routing is to IP. IISP creates a routed network by using defined static routes between switches. To get from one point to another, an IISP network looks at the NSAP (Network Service Access Point) address of a packet and then routes the cell accordingly. IISP uses VCIs (virtual channel identifiers) 32 to 255 on VPI (virtual path identifier) 0 as the actual "trunks" on which to carry the data.

To create the connections, you need to set up each side of a trunk as being user or network. These designations are assigned arbitrarily: Both sides have signaling capability and have no meaning other than names for "one side" and "the other side" (see "IISP Setup," above).

By using NSAP-based AESAs (ATM End-System Addresses), IISP matches the longest address prefix in a switch's static routing table to the destination address. Although IISP is very simple to set up in a small environment, it soon becomes as unwieldy as static routes in your enterprise IP router and has a limited ability to scale. And though creating very large routing tables and running thousands of routes are possible, the horror of managing and maintaining the tables should be left only to your worst enemy. The same router performance issues from large static IP routes will exist with large IISP routes, which is a drawback for larger networks.

You'll need to do serious planning to create the routing tables used by IISP, which can do only what it's told. Routes are created and used by their cost, with lowest cost routes being used first. If a switch determines that a route is down, the switch uses the next cost route. The problem with this method is that IISP has no way to avoid routing loops. It can determine when cells are in a loop but has no way of avoiding the loop or working around the loop until a better link with a lower cost becomes available. A valid path may be available, but an ATM switch running IISP will go for the first route in its table that meets the criteria.

The difficulty of designing routes for all possible catastrophes only increases with the number of switches and routes in the network. While IISP is used and works well in very small networks, it was never designed to be the end-all ATM routing protocol.

PNNI is similar to OSPF routing, which is commonly used for IP traffic. PNNI creates a route through the network based on current path information. PNNI comprises two protocols working together:

• The signaling protocol handles functions such as soft PVCs and crankback indications, and sets up the ATM connection on the path determined by the routing protocol.

• The routing protocol exchanges the hierarchical network topology between nodes, letting each switch in the network see enough of the big picture as is necessary to operate.