High-Availability Cabling Topologies

The standards place considerable emphasis on quality recommendations, but the aspects of high availability are not considered. In fact, the connections of the primary and secondary network levels act as single points of failure for all workplaces in the underlying levels. In the case of a backbone failure, complete departments lose the ability to communicate.

As the following example will show, a hierarchical star or tree topology produces a structured, powerful, and flexible approach for an SAP system infrastructure. However, you will see that the potential issues remain relative to high availability. This is an example of a network cabling structure built to support SAP systems with approximately 200 users at an automotive supply company in Germany.

The topology shown in Figure 9-7 is typical for a medium-sized manufacturing plant. The different departments are situated in several dedicated buildings spread over the property. The PC density ranges from very high within offices to low within the warehouses, but most buildings need at least some network connectivity. The network carries mission-critical data as well as email, web pages, and high volume CAD data. The cabling follows a layered, hierarchical star concept. PCs are connected to workgroup switches and hubs in the building cabling cabinets via dedicated copper cables. The cabling cabinets are connected to the data center with fiber cables (recommended). High-speed technologies deliver more or less appropriate bandwidth to the departments.

This topology is adequate from a performance point of view. However, it does contain quite a few network single points of failure. One failure at the lifeline, the main fiber cable between the data center and workgroup switch, and at least one complete department is unable to perform its task. Many companies have learned the hard way, for example, that even the loading dock can be mission critical. Without printed papers, no delivery truck leaves the property. Fortunately, the cabling infrastructures seldom fail, but when a problem does occur, repair cannot take a long time. It is more complicated than replacing a box. For this a high-availability solution is needed, at least for the network backbone.

Ring Topology

To avoid the single points of failure, the Fiber Distributed Data Interface (FDDI) technology provides a redundant cabling structure. One fiber pair transmits 100 Mbps shared between all users; a second pair is cold standby. In the case of a component failure, the double rings automatically wrap to a single ring of double length. Before the advent of switching technology, you would find this technology in all relevant SAP whitepapers recommended for the network backbone.

Because of higher costs for FDDI interface cards, it is common to connect only mission-critical servers directly to the fiber optic ring. FDDI bridges help connect end-user segments based on Ethernet or Token Ring technology to the rings. This configuration has a severe drawback: The FDDI bridges are single points of failure. Because all users connected to a failed FDDI bridge lose their connectivity, end-to-end high availability is not assured. FDDI is a shared technology with no migration path to higher bandwidths. Half of the cabling and ports are redundant and standby only, wasting potential network bandwidth. In addition, bridging FDDI to Ethernet or Token Ring causes latency for protocol conversion. Switched Fast and Giga-Ethernet have outperformed FDDI in terms of bandwidth but don't have FDDI's built-in failure tolerance. So, new methods must be found for providing fast and high-available networks.

Meshed Topology

Fully meshed network topologies provide the highest level of availability (see Figure 9-8). For wide area networks, proprietary switching technologies enable meshed networks. The backbones of many public WAN carriers deploy this technology for high availability. As described in Chapter 11, the availability of the network access points remains a problem.

Common LAN technologies do not allow multiple paths between devices. Meshed LANs are only possible with routers, special layer-3 switches, or ATM technology. This, however, raises the costs and complexity and impacts the transmission times as well.

The next section introduces a new, innovative cabling concept that brings together the best of both LAN worlds: the failure tolerance of FDDI topology and the benefits of the hierarchical star topology.

Cable Cluster—A High-Availability Cable Concept

Within the area of server technology, clustering is a well-known concept to protect mission-critical applications against the failure of a single component. In comparison to a cold standby solution, all servers within the cluster can be active. The failure of one server may only cause a reduction in the overall system performance, depending on how much the failed server was utilized.

Figure 9-9 shows an innovative cabling concept that provides a combination of the failure tolerance of FDDI topology and the benefits of the hierarchical star topology.

In this approach, a multi-fiber trunk cable circles the campus, terminated at both ends at the data center patch panel. At every cabling cabinet, the trunk cable is spliced. One (or more) fiber pairs are cut through and both ends are terminated at the local patch panel. This can be done easily by fusion welding prefabricated pigtails to the sliced cable. At least two workgroup switches are installed in each cabinet. The copper cables of the tertiary cabling are patched to these switches in such a way that no two user PCs with the same critical business functions are connected to the same workgroup switch. A disaster at any point of the cabling causes the loss of connectivity for some users, but the departments can still operate. Changing the patch cables provides a fast work around to bring the complete department online again until the defect is fixed.

One step closer to a high-availability configuration is the combination with HP's switch meshing. This technology allows up to four full duplex, Fast-Ethernet connections between two switches, each running different ways. This way, up to 800 Mbps are delivered to any workgroup switch (details regarding this are explained later). A disaster at any point of the cabling simply causes a loss of bandwidth, but no loss of connection.

Figure 9-10 is an example using a failure-tolerant cabling cluster configuration, implemented at a company with approximately 500 users. Two trunk fiber optic cables span the perimeter of the property. Those departments defined as mission critical are connected to the clock wise, as well as to the counterclockwise running trunk. The two backbone switches and the mission-critical servers in the data center are divided by a solid physical wall used for fire retardation. In addition to high availability, at least 400 Mbps (up to 800 Mbps maximum) network performance bandwidth is dedicated to the departments. The investment for this high-availability cabling solution was approximately 10% more than classical single-points-of-failure structures.

This is an easy and relatively low cost way, compared with the full SAP project implementation costs, to add significantly to the overall availability of the mission-critical system, while improving the network performance and response time seen at the desktop.