Exchange Server 2013 : Defining a Highly Available Messaging Solution - Defining Availability

The definition of availability and how that availability is measured is one of the key factors in the choice of technology you will use to implement in your design. One of the critical factors to establish is how the availability of the system will be measured and what the context for that measurement will be. We will use the following formula to calculate availability levels:

percentage availability = (total elapsed time – sum of downtime)/total elapsed timeM

Using the example of 99.9 percent availability, which is often noted as “three nines” availability, this translates to the system suffering downtime of 8.76 hours on an annual basis. That may not sound like a lot; if the measure is changed to three nines per day, this translates to 1.44 minutes of downtime, which may or may not be enough time for a reboot to occur.

If you calculate the permutations from one nine to five nines, you will understand why you don't often hear of systems that are available beyond five nines, as shown in from Table 1. This depends, however, entirely on how the availability is measured.

TABLE 1: Availability by the nines

Defining Availability Components

One of the many things you'll do during requirement elicitation is to help the business define and understand the services that Exchange delivers in terms of Exchange availability. At minimum, Exchange availability is a superset of dependency services. These services may be classified as follows:

• Client access
• Email transport
• Email storage

For the sake of clarity, we will not define auxiliary services such as message hygiene third-party application integration, or the many other examples that come to mind. Assuming you are satisfied to continue with the three basic services listed, your criteria may also include that availability could be measured per service. The availability of a system consisting of several independent critical components is a product of the availabilities of each individual component. The product is calculated by multiplying together the availabilities of each individual component in the following manner:

In the case of three components, our equation becomes

Let us examine a hypothetical example of three nines availability across three hypothetical components, as shown in Table 2. Using Excel, you would list the three availability components and use the PRODUCT function to multiply the three numbers together. Or, you could simply multiply the first number by the second number and then by the third number to obtain the fourth number, or result. Since we have multiplied three numbers together to obtain a fourth number, we need to move our decimal to the left by four places, or divide by 10,000 (four zeroes), to obtain a number that may be expressed as a percentage availability. The result will look similar to Table 2.

TABLE 4.2: Component availability

Notice that the total availability is lower than the availability of any of the component pieces. In order to build a system with a particular stated resilience, the components' availability requires examination. This may include network, power, chassis, and storage availability to name just a few, because these are all dependent features in larger systems.

Figure 1 illustrates the interdependency of component systems. Note that each one of the boxes of components can carry its own availability measurement when calculating total system availability.

FIGURE 1 Interdependent systems Credit: Boris Lokhvitsky

Figure 1 clearly illustrates that it is extremely difficult to establish even theoretical total system availability when so many systems are interrelated. When agreeing on the resulting availability of the desired system, clarify the definition of availability, downtime, and scheduled downtime as it pertains to the business.

Scheduled system downtime classically is not included in how availability is measured—in our three nines example, 1.44 minutes per day does not allow for much action to be taken. However, if the measure is adjusted to per year, then 8.76 hours becomes much more plausible for system maintenance.