Prof. Lawrence Chung (email@example.com) is in Computer Science at the University of Texas at Dallas. He has been working in System/Requirements Engineering and System/Software Architecture. He was the principal author of the research monograph “Non-Functional Requirements in Software Engineering", and has been involved in developing “RE-Tools” (a multi-notational tool for RE) with Dr. Sam Supakkul, “HOPE” (a smartphone application for people with disabilities) with Dr. Rutvij Mehta, and “Silverlining” (a cloud forecaster) with Tom Hill and many others. He has been a keynote speaker, invited lecturer, co-editor-in-chief for Journal of Innovative Software, editorial board member for Requirements Engineering Journal, editor for ETRI Journal, and program co-chair for international events. He received his Ph.D. in Computer Science in 1993 from University of Toronto.
What are non-functional requirements (NFRs)?
NFRs colloquially have been called “-ilities” and “-ities”, since many words referring to NFRs end with “-ility” (e.g., usability, flexibility, reliability, maintainability) or “-ity” (e.g., security, integrity, simplicity, ubiquity). There are of course many other words that do not end with either “-ility” or “-ity”, such as performance, user-friendliness, power consumption, and esthetics, but still refer to NFRs.
Functional requirements (FRs), in contrast, are about functions, activities, tasks, etc. that may accept some input and produce some output.
Consider, for example, “add” (“+”) on a calculator which adds two numbers given as input and produces another number as output shown on the screen. Now suppose you type “2 + 3 =” now and the calculator shows “5”, but one year from now. In this case, the “add” on the calculator is functionally correct but non-functionally terrible, in particular, concerning performance.
As even in this simple example, a system which fulfills only functional requirements is often times not usable or even not useful.
- So, handle NFRs and handle them appropriately. Don’t spend time only on FRs.
The “soft” Characteristics of NFRs and how to deal with them:
NFRs are global, subjective, interacting and graded.
FRs, such as “The calculator shall offer an “add” function”, are local in the sense that they are specific to the particular functions and not applicable to other functions or globally to other systems, such as a “subtract” function or a banking system. However, NFR terms such as “performance” can be applied to many other functions and systems, such as a “subtract” function and a banking system, and also those parts of such a function and a system.
In contrast to FRs, NFRs are subjective in both their definitions and the manner they need to be met – some are more subjective than others. Concerning definitions, for example, usability may mean simplicity and the availability of many help facilities to some people, while the same may mean something different, such as minimal learning curve and fast response. Concerning the manner whereby NFRs are seen satisfactorily met also depends on the (perception of the) user. For example, the keyboard with tiny keys on a smartphone may be usable to young people but not to old people. Also, large keys may be good enough for some people in using a smartphone, but a context-sensitive help may additionally be needed for a smartphone to be considered usable for some other people.
- So, clarify the definitions of NFRs. Don’t assume they have unanimously agreeable definitions.
- So, operationalize NFRs. Don’t just leave them without how they can be met.
NFRs are also interacting with each other, either synergistically or antagonistically or both. For example, a heavy authentication mechanism, for the purpose of enhanced security, may be hurting usability. If it takes three different passwords, which have to be changed every month and should consist of at least one special character, one digit, one upper case character, one function key, etc., in order to get in the system, the user of the system is unlikely to feel that the system is user-friendly. Hence, a conflict between security and user-friendliness. But a heavy security mechanism may help prevent unauthorized people from entering fake data in the system, hence a synergy between the security and the accuracy of data.
- So, identify conflicts among NFRs. Don’t think you can do anything with them individually without any negative consequences.
- So, identify synergies among NFRs. This is how we get “the whole becoming bigger than the sum of its parts”.
NFRs are graded, in the sense that they are usually met to different degrees. For example, an “add” function may be seen to be very good, good, bad or very bad, concerning its performance or usability, and different ways to implement the add function may affect the function differently – e.g., fully positively, partially positively, fully negatively, and partially negatively.
- So, consider the degree of contributions between NFR-related concepts. Don’t simply think NFR-related concepts affect each other in a binary manner – either a complete satisfaction or dissatisfaction.
In a nutshell, NFRs cannot be defined or met absolutely in a clear-cut sense, i.e., soft.
- So, satisfice NFRs. Don’t think NFRs can be satisfied absolutely, whatever the term “absolutely” might mean.
Product- vs. process-oriented approaches:
In science, objective measurements are important. But, are we mature enough to do that in system/software engineering? Also, consider:
“Not everything that can be counted counts, and not everything that counts can be counted.” [Albert Einstein]
According to this wisdom, it seems we need to measure important NFRs and only if we can. For example, you wouldn’t say “I love you 8 love units tonight”. It also seems that we need to shift our emphasis from measuring how well NFRs are met by a system/software artifact to how to handle NFRs during the process of developing the artifact in such a manner that the resulting artifact can be measured well.
- So, treat NFRs as (soft)goals to satisfice. Do not repeat developing, scraping, and redeveloping a system, if they do not meet the expected NFRs, until a good system is finally produced.
Rationalize decisions using NFRs:
A (functional) problem may be solvable in many different ways. For example, break entries may be stopped by having a security guard, a housedog, a fortified gate, a home security software system, etc. Similarly, a (functional) goal may be achievable in many different ways also. Which one do we decide to choose and how? We use NFRs as the criteria in making the decision on selecting among the (functional) alternatives. Furthermore, NFRs treated as softgoals naturally lead to the consideration of such alternatives, among which a selection is made.
- So, use NFRs as softgoals in exploring alternatives and also as the criteria in selecting among them.
How many NFRs are out there?
There can be many FRs. How about NFRs? If we go through a reasonably comprehensive dictionary and consider how many words can end with “-ility”, “-ity”, “-ness”, etc., this might give a hint. It’s not in the order of tens or even hundreds, but potentially thousands and tens of thousands. Alas, but we have resource limitations – a limited amount of time and money, our memory and reasoning capabilities, etc.
- So, prioritize NFRs and their operationalizations throughout the softgoal-oriented process. Don’t simply claim “Our system satisfies all the possible NFRs and absolutely”.