In a recent post titled “Architectural Drivers”, Davy Brion listed the factors that influence his architectural choices. Additionally, he makes a number of important observations, such as the wisdom of avoiding unnecessary complexity and the fact that there is no absolute “right” or “wrong”, only trade-offs. His efforts inspired me to put together my own list of drivers, several of which are the same or at least related to his. Additionally, I include some discussion of how they affect the architecture of a given system or solution.
Functionality being the first consideration should not come as a surprise since the name of the site is “Form Follows Function”. Are the user interactions task-oriented processes, basic CRUD operations, or a combination of both? Are those interactions real-time, asynchronous, or a mixture? Is the system stand-alone, server-based, peer to peer, or a hybrid? How much, if any, of the desired functionality is available from third party providers?
The requirements that describe the system’s behavior should be a major determinant of the solution’s application architecture, data architecture, and infrastructure architecture. It is important to look at the requirements in a holistic manner. Although those of the current iteration of the current release will be foremost in mind, some thought should be given to those that will be coming later to avoid painting yourself into a corner. It is also important to understand that the functional requirements will also influence many quality of service requirements. For example, the nature of the processes will affect the authorization scheme, the ability to process asynchronously will affect performance and scalability, etc.
The characteristics of the data to be managed will be a major driver of the architecture. Does the system deal with structured data, unstructured data, or a mixture? If the data is structured, does the functionality infer the need for a relational database or one of the NoSQL variants? Is the data sensitive (either wholly or partly), requiring extra security? How many records/documents will the system handle per day? How large are the units of data? Will the volume of data grow linearly or exponentially? Is there a need for archiving or near-line storage? Does the data lend itself to caching, either in whole or in part? Will there be a need to import data from a prior version of the system or from other systems to be replaced by this one?
This driver will have a major impact on the data architecture, infrastructure architecture and application architecture. Additionally, the characteristics of the data will affect a wide range of non-functional requirements: availability, backup requirements, certification, compliance, disaster recovery requirements, performance, response time, scalability, and security.
Who uses/will use a system significantly influences its architecture. Is the system intended for users internal to an organization, its customers, vendors, or some combination of these? Does the system interact directly with users, provide services to other applications, or both? Does the system have to support multiple platforms, either as a host or client (e.g. mobile)? Is accessibility an issue?
The answers to these questions will particularly impact quality of service concerns such as interoperability, portability, security, and usability. Likewise, this aspect is a significant driver of the infrastructure architecture.
Question around how heavily the system will be used are as significant as the issue of who will be using it. Numbers of concurrent users and operation counts will have considerable impact. Another consideration will be the pattern of that usage: business hours versus around the clock. Will it be steady or spiky? If there are spikes, will they occur on a daily, weekly or monthly basis?
This driver will primarily affect quality of service requirements such as performance, response time, and scalability. If the usage pattern is twenty-four hours a day or nearly so, this will also affect maintenance and backup considerations too. The impact on infrastructure architecture is obvious, but it will also drive application and data architecture as well.
The nature of the system in relation to the enterprise, ranging from mission critical to minor, is a key architectural driver. Gartner’s application classes, as described in my previous post “Application Lifecycle Management Revisited”, may be useful in classifying the business priority of a system as well. Time to market can also be a component of this driver, particularly for systems of innovation. The business owner’s tolerance for risk will also fall under this category, with greater tolerance leading to less emphasis on the associated quality of service requirements. A system of innovation with a business owner having a high tolerance for risk will be a good candidate for quickly accumulating technical debt.
Business priority will drive application architecture, data architecture, and infrastructure architecture. Non-functional requirements associated with this driver include availability, backup requirements, disaster recovery requirements, maintainability, performance, response time, scalability, security, and supportability.
Regulatory and Legal Obligations
Regulatory and legal compliance can have as wide ranging an impact on application, data and infrastructure architecture as functionality and business priority. The salient feature of this driver, however, is that there will be little flexibility around it. If a regulation applies, then the solution complies (or else). This is not to say that there will be no gray areas with this driver, just that they will generally be questions of whether the system falls under the jurisdiction of a particular law or regulation rather than whether or not to comply if it does. HIPAA and Sarbanes-Oxley (SOX) are but two examples of the myriad potential legal obligations that could apply. Other examples of legal issues include licensing compliance (for both commercial and Open Source components and code), copyright, trademark, and patent issues.
Common quality of service requirements associated with this driver include accessibility, auditability, certification, compliance, and security. This is not to say, however, that others could not be impacted by legal and regulatory obligations.
If an enterprise has formal architectural standards, obviously they will affect application, infrastructure, and data architecture, including the full range of non-functional requirements. Absence of formal standards, however, should not be mistaken for absence of constraint. The existing environment (e.g. hardware and operating systems, development team background, support team background, etc.) will act as this type of driver in the absence of formal standards or where such formal standards are ignored.
Non-functional requirements associated with this driver will typically include accessibility, extensibility, maintainability, security, supportability, testability, and usability.
Arguably, audit requirements could have been lumped under functionality, but I listed it separately because of the variety of sources that these requirements can come from. Laws, regulations, the organization’s internal auditors, and even the business users may require tracking system activity. Audit information can also be valuable in determining usage volume and patterns, making operations another potential stakeholder for this driver.
Audit requirements will primarily drive application and data architecture. Obviously, auditability is the primary non-functional requirement affected, but performance, response time, scalability, security, and supportability can all be impacted as well.
Like audit requirements, reporting considerations could be considered a sub-topic of functionality. I list it separately, however, because this driver can be both internal and external to the solution. Reporting that is integral to the solution can affect application, data, and infrastructure architecture (e.g. maintaining a dedicated reporting database to offload reports traffic is a common tactic). Enterprise reporting considerations can likewise impact the architecture of the solution, whether because of the need to support ETL jobs or to participate in master data management, as a supplier, consumer, or both.
This driver can impact quality of service requirements such as auditability (in situations where even reads must be logged), availability, backup requirements, performance, response time, scalability, and security.
Dependencies and Integrations
In “Managing Dependencies”, I outlined a number of ways in which the application architecture is affected by dependency management practices (see “Dependency Management – Anti-Patterns as Art from The Daily WTF” for some humorous, but true examples of how poor dependency management can affect architecture). Likewise, “Using Extension Methods for Message Transformation” and “Strict Versioning for Services – Applying the Open/Closed Principle” cover some of the effects of integrations (using services) on application architecture. Integrations and dependencies may also affect infrastructure (e.g. when connections come from systems external to your network or when using file drops) and data architecture (e.g. when using shared databases).
Non-functional requirements potentially impacted by dependencies and integrations include availability, disaster recovery requirements, interoperability, performance, response time, scalability, security, and supportability. Extensibility could also be potentially affected by this driver. A key consideration in the “Using Extension Methods for Message Transformation” and “Strict Versioning for Services – Applying the Open/Closed Principle” posts is structuring the system to avoid situations where two or more teams must conduct parallel development and releases to avoid breaking integrations.
I list this driver separate from business priority because it’s a fact of life that funds may be limited, no matter how critical the system may be. Salaries for development and operations staff; licenses for operating systems, database servers, and other dependencies; servers, storage appliances, network hardware and capacity all cost money (and this is far from a comprehensive list). These costs must be accounted for to properly understand their impact on the solution as a whole given that all aspects of the architecture are vulnerable to this driver.
While this is the last driver in my list, it is by no means the least. New development may be subject to fewer constraints than changes to existing systems, but it isn’t without issues. Existing architecture can be both a limit to your freedom and also a base to build on.
In my opinion, these twelve drivers provide a framework for making architectural decisions for technology solutions, whether built, bought, or rented. It is important not to infer the importance of any one driver from its position in the list; that ranking will be a function of the priorities of the business owner(s) plus the requirements of the enterprise as a whole. Establishing the priorities will assist in balancing the conflict between favoring the simplest possible solution and providing sufficient flexibility to grow and change as time goes on. Additionally, these drivers provide a framework for ensuring that decisions are driven by need and not fads or dogma.