3.11 Requirements for Predictable and Guaranteed Performance

Along with determining performance requirements, thresholds, and limits, we must also consider whether there are any requirements for predictable or guaranteed performance. You may recall from Chapter 1 that predictable service (of which performance is a part) requires some predictability and reliability, more than best effort, whereas guaranteed service has accountability as part of its specification. The performance requirements for both of these service types are much more strict than the standard best-effort service. Thus, when specifying predictable and/or guaranteed performance for a network, you will want to be certain that they really need or want it and that they are aware of the costs (e.g., financial, personnel, intellectual, possibly schedule) to implement and maintain it.

3.11.1 Requirements for Predictable Performance

Several of the examples we have discussed in this chapter may be considered predictable. Depending on how important the users and applications are to the success of that organization, their requirements may need to be predictable, requiring more support for their traffic flows. This would mean that, in the network architecture and design, their flows are handled differently from best effort. How that will be done will be covered later in this book, but for now, it is sufficient to be able to identify and specify such predictable performance requirements.

There have been several indicators of predictable performance so far in this book. We have talked about mission-critical, rate-critical, real-time, and interactive applications. They indicate predictable or guaranteed capacity, delay, and RMA, respectively. We have also talked about developing general and environment-specific thresholds and limits to separate low and high performance for multitier performance networks. Those applications that are characterized as high performance are also candidates for predictable service. In fact, the threshold between low and high performance can often be the threshold between best-effort and predictable service.

Therefore, whether or not you specify performance requirements as predictable is based on the following:

Determining whether the application is mission-critical, rate-critical, real-time, or interactive
Determining whether there are any environment-specific performance thresholds or limits
Applying general thresholds and limits, if necessary
Discussing the results with your customer(s) (e.g., users, management) to agree on which requirements should be considered predictable

3.11.2 Requirements for Guaranteed Performance

Guaranteed performance requirements are the most strict requirements for the network—not necessarily in terms of the level of performance but in terms of what is needed to support (or guarantee) that performance to the device, application, or user. As with predictable requirements, guaranteed requirements are indicated when mission-critical, rate-critical, real-time, and interactive applications are identified, as well as when high-performance (multitier) requirements are identified.

What makes a performance requirement guaranteed is the degree to which that requirement needs to be supported in the network. Support for guaranteed requirements has to have accountability built into it. This means:

There must be some agreement, usually between users of the application and providers of the network service, that supports the application, about the following:
1. What the performance requirements are
2. When and where they apply
3. How they will be measured and verified
4. What happens when a requirement is not met
This agreement may be in the form of an SLA, commonly used by service providers.
The requirements must be considered end-to-end, between source and destination. You can find more on this in Chapter 4.

The factors used to determine guaranteed performance requirements are the same as those for predictable performance requirements, with particular emphasis on the last factor (discussing the results with your customer). Getting agreement from your customer on which performance requirements must be guaranteed is especially important, for the impact on resources can be substantial.

As we will see, how you specify requirements here will have a serious impact on the network architecture and design to follow. Therefore, the more effort you put into this, the better your resulting architecture and design will be.