Network Computingwww.networkcomputing.com

More important, because virtual interfaces are defined in software and not tied to physical wires, they can be reproduced easily, and remotely, on a spare router in the event of a router hardware failure. All that's required is that you load the afflicted router's configuration, which you previously saved to a TFTP server, into a spare router that is connected to the ATM cloud.

Another advantage to software-defined interfaces is that you can easily fine-tune router load and link bandwidth by moving virtua l interfaces among routers in response to changing traffic patterns.

Typically, all you have to do is telnet to both routers, copy the interface configuration from one router, paste it into the other, and delete the virtual interface from the first router. This takes about a minute. Because you can make changes like this more frequently relative to traditional router setups, you can better utilize your router and switching investment. Since upgrading, we have been able to keep the CPU utilization on our campus routers to within 10 percent of each other.

Finally, virtual interfaces let you separate legacy protocols from strategic ones. In the "One-Armed Router Approach" diagram (below left), notice that Router A runs IP only. Router B runs IPX and AppleTalk. Each router is a member of four ELANs that service Subnets 1 through 4. Router A provides these subnets with connectivity for IP, while Router B prov ides IPX/AppleTalk routing for these subnets.

By separating the protocols, you aggressively p ursue strategic directions, such as Resource Reservation Protocol (RSVP), IP multicast, IP optimized routers, IPv6 and other IP enhancements, without having to regression-test legacy services every time you upgrade a router. In effect, you have quarantined the legacy protocols and can simply let them "die on the vine" with a known stable code revision.

Balancing Act In Ethernet networks, you can provision redundant links between two points. However, the Spanning Tree protocol, which ensures a loop-free topology, precludes load-balancing traffic across these redundant links. Proprietary methods for load-balancing traffic over multiple Ethernet links exist, but they are limited to trunks between two switches or between a switch and a NIC. And, you normally have to disable the Spanning Tree safety net for this to work.

ATM, on the other hand, can load-balance traffic on a call-by-call basis across as many parallel links as are provisioned. Private Network-to-Node Interface (PNNI), ATM's routing protocol, uses source routing, so routing loops caused by parallel paths are not a possibility.

In addition, PNNI leaves a lot of room for vendors to add internal enhancements to ATM switch software while still presenting a standards-based interface to the outside world. For instance, some vendors have tailored their switches to load-balance in accordance with several different algorithms. But, in practice, we have found that a round-robin balancing algorithm does a reasonable job of balancing load for Unspecified Bit Rate (UBR) traffic, such as LAN Emulation (LANE), across parallel links between two switches.