The Influence of Network on Availability

In the beginning of this book we explained what availability means for the servers and storage devices for a mission-critical system like SAP. Various methods of making these systems more available for an interruption-free business operation were presented, along with some recommendations. However, availability is not an issue restricted to the data center. The connections to the computing power provided by the data center must also be reliable and highly available to the users, even those in more remote locations.

To cite an example of the importance of network availability, if the complete sales department loses connectivity to the SAP system because of damage to a simple network cable, no orders can be processed in spite of a highly available server cluster in the data center. In a more and more globalized economy, it is quite common, even for smaller mid-market companies, to distribute sales offices, as well as factories and distribution centers, throughout the world. All these subsidiaries have to be connected to the central SAP systems at headquarters. However, if no delivery forms can be printed at a production unit abroad because of unreliable wide area network connections, no truck will leave the loading dock toward the customer in spite of a 24-hour operable data center.

As the integration of supply chains between enterprises grows stronger, the necessity of highly available connections, even on the Internet as a basis for mySAP.com, gains in importance. For a just-in-time production business, it is essential to have highly available connections between the SAP systems of the partner companies. What counts in a highly integrated SAP system environment is the end-to-end availability. Multi-tier systems, like the different SAP applications, must be viewed as a whole.

Early on in the system design planning stage, the degree of availability of the business function or process for the different user groups should be defined. From this definition, the special demands on the different areas of data networks and links can be derived.

Because of the strategic role network infrastructures play, reliability issues create a lot of interest in the early stages of development. The Advanced Research Projects Agency (ARPA), then part of the U.S. Department of Defense, developed the ARPAnet with the ultimate goal of providing connectivity between command posts even under the conditions of a nuclear war. This goal was reached primarily by meshing. The necessity of automatically finding alternative routes through a highly dynamic mesh of links caused the advent of the router. A few generations later, the ARPAnet is now considered the Internet's ancestor.

As shown earlier in this chapter, the data streams between SAP server systems and workplace computers pass a multitude of intermediate devices and links. Each component has a distinct level of availability. The availability of the system as a whole is given by multiplication of the availability of any single component within the link (as long as there is no redundancy). This results in the effect that the availability of the system as a whole is less than the availability of each individual component.

To demonstrate the effect of the whole system availability problem, we will use a common example. The PC workplaces in a production plant abroad, or in another region, have to be connected to the central SAP systems in the enterprise data center. The PCs, of course, are first connected to the local network. The local network is linked by a router and a leased line to the Frame Relay network of an international telecommunication service provider (Telco). Using another leased line and router creates the link to the local network in the data center, where the SAP systems reside (see Figure 7-5).

Given a highly available server infrastructure in the data center, with a Unix system cluster configuration, the availability level of the system may reach 99.99%—less than an hour of unplanned downtime per year. For a best case estimate of the availability of the infrastructure, we will neglect the cabling and the availability of the PCs. The availability of the local network, as well as the routers, should be in the range of Unix servers without fail over (99.9%). The availability of the meshed network infrastructure of the telecommunication service provider has been engineered to achieve 99.999% uptime in most countries. For the leased line between customer locations and the access points to the Frame Relay provider (point-of-presence; POP), the German Telecom states an availability of 99.8%, for example. This is excellent compared with 95% known for PCs.

The availability of a network system as a whole is the product of the availability of the components within the transmission chain. This way, you will receive about 99.19%, or at minimum three days of unplanned downtime per year for an end-to-end connectivity between data center to a remote site. From practical experiences with network infrastructures similar to our example, the network outages last about four hours each and occur on the average nine times a year (experiences obviously vary throughout different regions of the world). If a period of four hours without connectivity to mission-critical applications is acceptable, then just about anything is OK. But let us imagine it's the order entry department or the product shipping dock in a just-in-time production facility; then your business is in trouble. A recent survey revealed the average loss across many industries is $330,000 per outage.

For situations where business-critical mySAP.com systems need end-to-end availability around the clock, we will introduce innovative high-availability approaches for all parts of a network infrastructure in the related chapters.