Evaluating a Software Architecture
- Why Evaluate an Architecture?
- 2 When Can an Architecture Be Evaluated?
- 3 Who's Involved?
- 4 What Result Does an Architecture Evaluation Produce?
- 5 For What Qualities Can We Evaluate an Architecture?
- 6 Why Are Quality Attributes Too Vague for Analysis?
- 7 What Are the Outputs of an Architecture Evaluation?
- 8 What Are the Benefits and Costs of Performing an Architecture Evaluation?
- 9 For Further Reading
- 10 Discussion Questions
Marry your architecture in haste and you can repent in leisure.
from a keynote address: And Very Few Lead Bullets Either
How can you be sure whether the architecture chosen for your software is the right one? How can you be sure that it won't lead to calamity but instead will pave the way through a smooth development and successful product?
It's not an easy question, and a lot rides on the outcome. The foundation for any software system is its architecture. The architecture will allow or preclude just about all of a system's quality attributes. Modifiability, performance, security, availability, reliabilityall of these are precast once the architecture is laid down. No amount of tuning or clever implementation tricks will wring any of these qualities out of a poorly architected system.
To put it bluntly, an architecture is a bet, a wager on the success of a system. Wouldn't it be nice to know in advance if you've placed your bet on a winner, as opposed to waiting until the system is mostly completed before knowing whether it will meet its requirements or not? If you're buying a system or paying for its development, wouldn't you like to have some assurance that it's started off down the right path? If you're the architect yourself, wouldn't you like to have a good way to validate your intuitions and experience, so that you can sleep at night knowing that the trust placed in your design is well founded?
Until recently, there were almost no methods of general utility to validate a software architecture. If performed at all, the approaches were spotty, ad hoc, and not repeatable. Because of that, they weren't particularly trustworthy. We can do better than that.
This is a guidebook of software architecture evaluation. It is built around a suite of three methods, all developed at the Software Engineering Institute, that can be applied to any software-intensive system:
- ATAM: Architecture Tradeoff Analysis Method
- SAAM: Software Architecture Analysis Method
- ARID: Active Reviews for Intermediate Designs
The methods as a group have a solid pedigree, having been applied for years on dozens of projects of all sizes and in a wide variety of domains. With these methods, the time has come to include software architecture evaluation as a standard step of any development paradigm. Evaluations represent a wise risk-mitigation effort and are relatively inexpensive. They pay for themselves in terms of costly errors and sleepless nights avoided.
Whereas the previous chapter introduced the concept of software architecture, this chapter lays the conceptual groundwork for architectural evaluation. It defines what we mean by software architecture and explains the kinds of properties for which an architecture can (and cannot) be evaluated.
First, let's restate what it is we're evaluating:
The software architecture of a program or computing system is the structure or structures of the system, which comprise software components, the externally visible properties of those components, and the relationships among them. [Bass 98]
By "externally visible" properties, we are referring to those assumptions other components can make of a component, such as its provided services, performance characteristics, fault handling, shared resource usage, and so on. The intent of this definition is that a software architecture must abstract some information about the system (otherwise there is no point looking at the architecturewe are simply viewing the entire system) and yet provide enough information to be a basis for analysis, decision making, and hence risk reduction (see the sidebar What's Architectural?).
The architecture defines the components (such as modules, objects, processes, subsystems, compilation units, and so forth) and the relevant relations (such as calls, sends-data-to, synchronizes-with, uses, depends-on, instantiates, and many more) among them. The architecture is the result of early design decisions that are necessary before a group of people can collaboratively build a software system. The larger or more distributed the group, the more vital the architecture is (and the group doesn't have to be very large before the architecture is vital).
One of the insights about architecture from Chapter 1 that you must fully embrace before you can understand architecture evaluation is this:
Architectures allow or preclude nearly all of the system's quality attributes.
Sooner or later everyone asks the question: "What's architectural?" Some people ask out of intellectual curiosity, but people who are evaluating architectures have a pressing need to understand what information is in and out of their realm of concern. Maybe you didn't ask the question exactly that way. Perhaps you asked it in one of the following ways:
What is the difference between an architecture and a high-level design?
Are details such as priorities of processes architectural?
Why should implementation considerations such as buffer overflows be treated as architectural?
Are interfaces to components part of the architecture?
If I have class diagrams, do I need anything else?
Is architecture concerned with run-time behavior or static structure?
Is the operating system part of the architecture? Is the programming language?
If I'm constrained to use a particular commercial product, is that architectural? If I'm free to choose from a wide range of commercial products, is that architectural?
Let's think about this in two ways.
First, consider the definition of architecture that we quoted in Chapter 1 of this book. Paraphrasing: A software architecture concerns the gross organization of a system described in terms of its components, their externally visible properties, and the relationships among them. True enough, but it fails to explicitly address the notion of context. If the scope of my concern is confined to a subsystem within a system that is part of a system of systems, then what I consider to be architectural will be different than what the architect of the system of systems considers to be architectural. Therefore, context influences what's architectural.
Second, let's ask, what is not architectural? It has been said that algorithms are not architectural; data structures are not architectural; details of data flow are not architectural. Well, again these statements are only partially true. Some properties of algorithms, such as their complexity, might have a dramatic effect on performance. Some properties of data structures, such as
whether they need to support concurrent access, directly impact performance and reliability. Some of the details of data flow, such as how components depend on specific message types or which components are allowed access to which data types, impact modifiability and security, respectively.
So is there a principle that we can use in determining what is architectural? Let's appeal to what architecture is used for to formulate our principle. Our criterion for something to be architectural is this: It must be a component, or a relationship between components, or a property (of components or relationships) that needs to be externally visible in order to reason about the ability of the system to meet its quality requirements or to support decomposition of the system into independently implementable pieces. Here are some corollaries of this principle:
Architecture describes what is in your system. When you have determined your context, you have determined a boundary that describes what is in and what is out of your system (which might be someone else's subsystem). Architecture describes the part that is in.
An architecture is an abstract depiction of your system. The information in an architecture is the most abstract and yet meaningful depiction of that aspect of the system. Given your architectural specification, there should not be a need for a more abstract description. That is not to say that all aspects of architecture are abstract, nor is it to say that there is an abstraction threshold that needs to be exceeded before a piece of design information can be considered architectural. You shouldn't worry if your architecture encroaches on what others might consider to be a more detailed design.
What's architectural should be critical for reasoning about critical requirements. The architecture bridges the gap between requirements and the rest of the design. If you feel that some information is critical for reasoning about how your system will meet its requirements then it is architectural. You, as the architect, are the best judge. On the other hand, if you can eliminate some details and still compose a forceful argument through models, simulation, walk-throughs, and so on about how your architecture will satisfy key requirements then those details do not belong. However, if you put too much detail into your architecture then it might not satisfy the next principle.
An architectural specification needs to be graspable. The whole point of a gross-level system depiction is that you can understand it and reason about it. Too much detail will defeat this purpose.
An architecture is constraining. It imposes requirements on all lower-level design specifications. I like to distinguish between when a decision is made and when it is realized. For example, I might determine a process prioritization strategy, a component redundancy strategy, or aset of encapsulation rules when designing an architecture; but I might not actually make priority assignments, determine the algorithm for a redun-dant calculation, or specify the details of an interface until much later.
In a nutshell:
To be architectural is to be the most abstract depiction of the system that enables reasoning about critical requirements and constrains all subsequent refinements.
If it sounds like finding all those aspects of your system that are architectural is difficult, that is true. It is unlikely that you will discover everything that is rchitectural up front, nor should you try. An architectural specification will evolve over time as you continually apply these principles in determining what's architectural.
2.1 Why Evaluate an Architecture?
The earlier you find a problem in a software project, the better off you are. The cost to fix an error found during requirements or early design phases is orders of magnitudes less to correct than the same error found during testing. Architecture is the product of the early design phase, and its effect on the system and the project is profound.
An unsuitable architecture will precipitate disaster on a project. Performance goals will not be met. Security goals will fall by the wayside. The customer will grow impatient because the right functionality is not available, and the system is too hard to change to add it. Schedules and budgets will be blown out of the water as the team scrambles to back-fit and hack their way through the problems. Months or years later, changes that could have been anticipated and planned for will be rejected because they are too costly. Plagues and pestilence cannot be too far behind.
Architecture also determines the structure of the project: configuration control libraries, schedules and budgets, performance goals, team structure, documentation organization, and testing and maintenance activities all are organized around the architecture. If it changes midstream because of some deficiency discovered late, the entire project can be thrown into chaos. It is much better to change the architecture before it has been frozen into existence by the establishment of downstream artifacts based on it.
Architecture evaluation is a cheap way to avoid disaster. The methods in this book are meant to be applied while the architecture is a paper specification (of course, they can be applied later as well), and so they involve running a series of simple thought experiments. They each require assembling relevant stakeholders for a structured session of brainstorming, presentation, and analysis. All told, the average architecture evaluation adds no more than a few days to the project schedule.
To put it another way, if you were building a house, you wouldn't think of proceeding without carefully looking at the blueprints before construction began. You would happily spend the small amount of extra time because you know it's much better to discover a missing bedroom while the architecture is just a blueprint, rather than on moving day.