![]() ![]() Hardcore ATM Switches for the WAN Routing is another example where vendors have their own, non-standard method. Each vendor makes use of common cost metrics in determining the right path through the network. While the ATM Private Network-to-Network Interface (PNNI) has been ratified by the ATM Forum, none of these products yet supports it. Nortel uses three different route determination algorithms, depending on the type of service and end points, using an Open Shortest Path First (OSPF)-like method for connectionless services. IBM uses a proprietary Spanning Tree algorithm to distribute topology information dynamically through the network. GDC uses static route tables built offline with a route generation tool. Note that this does not imply the absence of failover paths and dynamic rerouting, but simply means that new routes are not generated automatically by the network when a topology change occurs. The three vendors also take different approaches to the NMS . IBM's switch uses Common Management Information Protocol (CMIP) over TCP/IP to communicate with the NMS, a protocol found principally in carrier networks. Although the ATM Forum remains agnostic in the CMIP versus SNMP debate, most ATM NMS implementations are Simple Network Management Protocol (SNMP)-based. GDC's solution is based on SNMP, while Nortel uses proprietary methods and SNMP interfaces. Nortel's NMS is by far the most comprehensive of the three, though each vendor's approach has praiseworthy elements. IBM distributes NMS functions around the network, putting an OS/2-based mini-NMS right on each switch. It's possible to manage a small network (fewer than 10 nodes) right from the built-in Node Administration Station. IBM's NetView-based NMS is the only solution that tightly managed all aspects of switch configuration and addressing on a networkwide scale. GDC' s approach is more open and standards-based, though the NMS version it demonstrated is relatively crude and in need of an overhaul. For example, GDC's real-time performance monitoring capabilities did not approach those of the Nortel or IBM systems. Common Threads Architecturally, the boxes share some common elements. All switches use a control processor to deal with configurations and overall management of the box. All have optional clock cards, designed to accurately hold synchronization from a truer reference source such as a carrier line. Both IBM and Nortel's clocks are rated at a Stratum-3 level (that is, drift of less than 4.6x10-6 units per year). GDC's are slightly less accurate than Stratum-3. Across products, GDC offers the smoothest upgrade path. There are five models in the APEX line, all with the same interface adapters and management platform. At the low end lies the APEX MAC 1 media access concentrator, a unit with no switching at all among ports. With the addition of a switch fabric module, the MAC 1 becomes a switch. The high-end APEX DV 2 has 19 slots. This integration, along with GDC's aggressive support of ATM Forum standards and good pricing, earn the APEX line Network Computing's Best Value award. All of the switches tested support high-availability options, including hot-swappable cards, fans and power supplies. All offer redundant central control processors. Actual interface port redundancy with automatic failover is available only on metallic interfaces from Nortel and only on optical interfaces from GDC. IBM does not offer automatic failover among interface ports. Each vendor takes a different approach to switch architecture. IBM and GDC place the switch fabric on the backplane (or midplane) of the switch, preallocating a slice of bandwidth for each card. With current port densities, traffic can't be blocked at the switch fabric. Alternatively, Nortel puts the fabric on each line card, using one or more passive buses to deliver traffic between cards, thus requiring s ome traffic management to avoid contention for the bus. The current maximum port speed of this class of switches is 155 Mbps (OC-3c), reflecting their use in WAN environments; in contrast, OC-12 interfaces are common on campus switches. While most port interface types are supported by all vendors, Nortel's per card densities were generally higher than the competition's (see "WAN Switch Features" on page 54). Hardware is important, but software is what really matters. The Magellan Passport has the widest range of advanced services in this group. Nortel provides switched virtual circuits (SVCs) for both ATM and frame relay and ATM Early Packet Discard, and the switch can use inverse multiplexing over ATM to bond multiple T1 channels into a single pipe. The switch can map Variable Bit Rate (VBR) traffic into 12 priority levels. IBM's switch software trails the pack, offering no support for SVCs of any kind and no ability to interface to a separate ATM network (other than as a transparent service, appearing to the end switches as a single hop). IBM does not support ATM-to-frame relay service interworking, preventing frame rela y-connected sites from communicating with ATM-connected sites. Only Nortel and GDC accept traffic over LAN interfaces, with Nortel leading in interface types and routing protocol support. GDC does not support routing at all, but offers only RFC 1483 bridging. Nortel has both Resource Information Protocol (RIP) and OSPF built into the units. However, on a cost-per-port basis, the Magellan platform is an expensive routing engine. How much does this all cost and How we tested This Issues other Feature Storming the Castle By David Willis Updated October 8, 1997 |















