Network Computingwww.networkcomputing.com

MPLS works by propagating route information from the egress point of the network and working inward. When a packet arrives at the edge of an MPLS-routed backbone, it is assigned a tag that indicates the packet's ulti mate destination. The packet is then forwarded into the network, where routers can switch it to the desired destination, cutting down on packet-processing overhead. MPLS has the potential to decrease latency and congestion if it becomes widely used on the Internet backbone. To set the record straight, most enterprise shops will never see MPLS in LANs or WANs; ISPs and carriers will benefit most from this technology.

In addition, Newbridge, Siemens and 3Com have formed a partnership to further develop Newbridge's CSI (Carrier Scale Internetworking), a WAN bandwidth management technology targeted at the backbone carrier. From a high level, CSI appears to be a set of protocols, filters and software methods that allow switches, such as Newbridge's MainStreet Xpress line, to guarantee bandwidth across the wide area. This is accomplished by utilizing ATM's QoS features and by monitoring congestion across the network. CSI is another one of those technologies you'll hear more about, but probably will never see on the LAN backbone.

Gazing Into the Crystal Ball Our society is being enveloped by the information age. The Internet has changed the way we do business. TCP/IP is the driving protocol behind these networks, and the need for more bandwidth and more bandwidth management is acutely visible.

FastIP, Layer 3 switches, MPOA and NetFlow Switching are all vying for a place in your next-generation backbone networks. But each of these technologies has costs, both upfront and down the road. Likewise, the technology you pick will depend greatly on your existing base of hardware. ATM users can experience radical changes in their network with MPOA-enabled hardware. Non-ATM shops may have a hard time justifying ATM this late in the game, however; ATM just isn't being adopted by the enterprise community at large. Traditional Ethernet shops have a much more difficult decision.

And in the long run, ASIC-driven Layer 3 switches will most likely populate many of the wiring closets on campus with big iron providi ng legacy connectivity to the wide area. These products will enable gigabit-speed networking at a price that will make all other local area technologies obsolete.

Joel Conover can be reached at jconover@nwc.com.

How We Tested
To measure latency of the Layer 3 routing solutions presented by Cisco Systems, Newbridge Networks and 3Com Corp., we utilized Layer 3 SmartCards from Netcom Systems. Using Netcom's Smart Windows Layer 3 tests, we measured end-to-end latency of the solutions provided. The three vendors brought a variety of hardware to the table, but the latency test setup was always measured from a 10-Mbps Ethernet port on an edge device, through a central switch, and out a 10-Mbps Ethernet port on a third ed ge device (see "Latency Test Setup," above).

To compare throughput, we used high-performance Pentium Pro workstations and Ganymede Software's Chariot. We also used Windows networking and custom-written load-generation software to compare and contrast the throughput of the various test configurations. In the case of 3Com's FastIP, we found that throughput was limited only by the speed of the links between switches. MPOA (Multiprotocol Over ATM) yielded similar results, though throughput in an MPOA network also can be limited by the route processing power of the edge device.

More noticeable with MPOA was the call setup time associated with a new flow of traffic. Both Cisco's and Newbridge's MPOA solutions had short VC (virtual circuit) lifetimes, meaning that the connections between edge devices were torn down after a certain inactivity time. According to Cisco, its one-second teardown time minimizes the chances of unauthorized packets traversing the VC. Newbridge uses a 15-second time-out. It was our ob servation that MPOA may very well put higher demands on the signaling capacity of your ATM core switches because of the number of calls being set up and torn down in the network.

Finally, we connected these solutions to the backbone network of Network Computing's lab at the University of Wisconsin and verified connectivity via OSPF, where applicable. Both Cisco and Newbridge interoperated with no problems. 3Com's FastIP depends on the existing router in your network, so connectivity wasn't an issue.