Disciplined Agile

Early in the lifecycle, during the Inception phase, disciplined agile teams will invest some time in initial requirements envisioning and initial architecture envisioning. One of the issues to be considered as part of requirements envisioning is to identify non-functional requirement (NFRs), also called quality of service (QoS) or simply quality requirements. The NFRs will drive many of your technical decisions that you make when envisioning your initial architectural strategy. These NFRs should be captured someone and implemented during Construction. It isn’t sufficient to simply implement the NFRs, you must also validate that you have done so appropriately. In this blog posting I overview a collection of agile strategies that you can apply to validate NFRs.

A mainstay of agile validation is the philosophy of whole team testing. The basic idea is that the team itself is responsible for validating its own work, they don’t simply write some code and then throw it over the wall to testers to validate. For organizations new to agile this means that testers sit side-by-side with developers, working together and learning from one another in a collaborative manner. Eventually people become generalizing specialists, T-skilled people, who have sufficient testing skills (and other skills).

Minimally your developers should be performing regression testing to the best of their ability, adopting a continuous integration (CI) strategy in which the regression test suite(s) are run automatically many times a day.  Advanced agile teams will take a test-driven development (TDD) approach where a single test is written just before sufficient production code which fulfills that test.  Regardless of when tests are written by the development team, either before or after the writing of the production code, some tests will validate functional requirements and some will validate non-functional requirements.

Whole team testing is great in theory, and it is strategy that I wholeheartedly recommend, but in some situations it proves insufficient.  It is wonderful to strive to have teams with sufficient skills to get the job done, but sometimes the situation is too complex to allow that.  There are some types of NFRs which require significant expertise to address properly: NFRs pertaining to security, usability, and reliability for example.  To validate these types of requirements, worse yet even to identify them, requires skill and sometimes even specialized (read expensive) tooling.  It would be a stretch to assume that all of your delivery teams will have this expertise and access to these tools.

Recognizing that whole team testing may not sufficiently address validating NFRs many organizations will supplement their whole team testing efforts with parallel independent testing  .  With this approach a delivery team makes their working builds available to a test team on a regular basis, minimally at the end of each iteration, and the testers perform the types of testing on it that the delivery team is either unable or unlikely to perform.  Knowing that some classes of NFRs may be missed by the team, independent test teams will look for those types of defects.  They will also perform pre-production system integration testing and exploratory testing to name a few.  Parallel independent testing is also common in regulatory compliance environments.

From a verification point of view some agile teams will perform either formal or informal reviews.  Experienced agilists prefer to avoid reviews due to their inherently long feedback cycle, which increases the average cost of addressing found defects, in favor of non-solo development strategies such as pair programming and modeling with others.  The challenge with non-solo strategies is that managers unfamiliar with agile techniques, or perhaps the real problem is that they’re still overly influenced by disproved traditional theories of yesteryear, believe that non-solo strategies reduce team productivity.  When done right non-solo strategies increase overall productivity, but the political battle required to convince management to allow your team to succeed often isn’t worth the trouble.

Another strategy for validating NFRs code analysis, both dynamic and static.  There is a range of analysis tools available to you that can address NFR types such as security, performance, and more.  These tools will not only identify potential problems with your code many of them will also provide summaries of what they found, metrics that you can leverage in your automated project dashboards.   This strategy of leveraging tool-generated metrics such as this is a technique which IBM calls Development Intelligence and is highly suggested as an enabler of agile governance in DAD. Disciplined agile teams will include invocation of code analysis tools from you CI scripts to support continuous validation throughout the lifecycle.

Your least effective validation option is end-of-lifecycle testing, in the traditional development world this would be referred to as a testing phase.  The problem with this strategy is that you in effect push significant risk, and significant costs, to the end of the lifecycle.  It has been known for several decades know that the average cost of fixing defects rises the longer it takes you to identify them, motivating you to adopt the more agile forms of testing that I described earlier.  Having said that I still run into organizations in the process of adopting agile techniques that haven’t really made embraced agile, as a result still leave most of their testing effort to the least effective time to do such work.  If you find yourself in that situation you will need to validate NFRs in addition to functional requirements.

To summarize, you have many options for validating NFRs on agile delivery teams.  The secret is to pick the right one(s) for the situation that you find yourself in.  The DA toolkit helps to guide you through these important process decisions, describing your options and the trade-offs associated with each one.

Related Resources

• Agile Testing and Quality Strategies 
• Develop Initial Test Strategy
• Strategies for Capturing Quality Requirements
• Strategies for Implementing Quality/Non-Functional Requirements

Non-functional requirements, also known as quality of service (QoS) or technical requirements, are typically system-wide thus they apply to many, and sometimes all of your functional requirements.  Part of ensuring that your solution is potentially consumable each iteration is ensuring that it fulfill its overall quality goals, including applicable NFRs.  This is particularly true with life-critical and mission-critical solutions.  Good sources for NFRs include your enterprise architects and operations staff, although any stakeholder is a potential source for NFRs.

As your stakeholders tell you about functional requirements they will also describe non-functional requirements (NFRs).  These NFRs may describe security access rights, availability requirements, performance concerns, or a host of other issues as saw in my blog regarding initial architecture envisioning.  There are three basic strategies, which can be combined, for capturing NFRs: technical stories; acceptance criteria for individual functional requirement (such as stories); and an explicit list.

So what are the implications for implementing NFRs given the three previous capture strategies?    Although in the DAD book we make this sort of comparison via a table to improve consumability, in this blog posting I will use prose due to width constraints.  Let’s consider each one:

1. Technical stories.  The advantages of this approach are that it is a simple strategy for capturing NFRs and that it works well for solutions with a few NFRs or simple NFRs.  But, the vast majority of NFRs are cross-cutting aspects to several functional stories and as a result cannot be implemented within a single iteration.  This strategy also runs the risk of teams leaving NFRs to the end of the construction phase, thereby pushing technical risk to the end of the lifecycle where it is most difficult and expensive to address.
2. Acceptance criteria. This is a quality focused approach which makes the complexity of an individual functional requirement apparent, working well with test driven approaches to development.  NFR details are typically identified on a just in time (JIT) basis during construction, fitting in well with a disciplined agile approach.  But, because many NFRs are cross cutting the same NFR will be captured for many functional requirements.  It requires the team to remember and consider all potential NFR issues (see Figure in my previous posting) for each functional requirement.  You will still need to consider NFRs as part of your initial architecture efforts otherwise you risk a major rework effort during the Construction phase because you missed a critical cross-cutting concern).
3. Explicit list.  This strategy enables you to explore NFRs early in the lifecycle and then address them in your architecture.  The list can be used to drive identification of acceptance criteria on a JIT basis.  But, NFR documents can become long for complex systems (due to the large number of NFRs).  This can be particularly problematic when you have a lot of NFRs that are specific to a small number of functional requirements.  Teams lacking in discipline may not write down the non-functional requirements and trust that they will remember to address them when they’re identifying acceptance criteria for individual stories.

In most situations you should maintain an explicit list and then use that to drive identification of acceptance criteria as we’ve found that it’s more efficient and lower risk in the long run.  Of course capturing NFRs is only one part of the overall process of addressing them.  You will also need to implement and validate them during construction, as well as address them in your architecture.

An important issue which goes to NFRs such as consumability, supportability, and operability, is that of deliverable documentation.  At the start of the project is the best time to identify the required documentation that must be created as part of the overall solution.  This potentially includes operations manuals, support manuals, training materials, system overview materials (such as an architecture handbook), and help manuals to name a few.  These deliverable documents will be developed and kept up to date via the continuous documentation practice.

Related Posts

• Strategies for Capturing Quality Requirements
• Strategies for Verifying Quality Requirements
• Develop Initial Test Strategy

An important aspect of Disciplined Agile Delivery (DAD) is its explicit inclusion of an Inception phase where project initiation activities occur.  Although phase tends to be a swear word within the agile community, the reality is that the vast majority of teams do some up front work at the beginning of a project.  Some people will mistakenly refer to this effort this Sprint/Iteration 0 it is easy to observe that on average this effort takes longer than a single iteration (the 2009 Agile Project Initiation survey  found the average agile team spends 3.9 weeks in Inception and the November 2010 Agile State of the Art survey found that agile teams have Construction iterations of a bit more than 2 weeks in length).

Regardless of terminology, agile teams are doing some up front work.  Part of that initial work is identifying an initial technical architecture, typically via some initial architecture envisioning http://www.agilemodeling.com/essays/initialArchitectureModeling.htm.  Because your architecture should be based on actual requirements, otherwise you’re “hacking in the large”, your team will also be doing some initial requirements envisioning  http://www.agilemodeling.com/essays/initialRequirementsModeling.htm in parallel.  Your architecture will be driven in part by functional requirements but more often the non-functional requirements, also called quality of service (QoS) or simply quality requirements.  Some potential quality requirements are depicted in the figure below (this figure is taken from the Disciplined Agile Delivery book but was first published in Agile Architecture Strategies ).

Some architects mistakenly believe that you need to do detailed up front modeling to capture these quality requirements and then act upon them.  This not only isn’t true it also proves to be quite risky in practice, see my discussion about Big Modeling Up Front (BMUF)  for more details.  Disciplined agilists instead will do just enough initial modeling up front and then address the details on a just-in-time (JIT) basis throughout construction.  Of course it’s important to recognize that just enough will vary depending on the context of the situation, teams finding themselves at scale will need to do a bit more modeling than those who don’t.  It’s also important to recognize that to address non-functional requirements throughout construction that you need to have more than just architectural modeling skills.  This topic will be the focus of my next blog posting in this series.

Recently at the Scott W. Ambler + Associates site we received a series of questions from someone who wanted to better understand how architecture issues are addressed on agile project teams.  It seemed to me that the questions were sufficiently generic to warrant a public response instead of a private one.  So, over the next few days I’m going to write several blog postings here to address the issues that were brought up in the questions.  It’s important to note that I will be answering from the point of view of Disciplined Agile Delivery (DAD), and not agile in general.  Other agile methods may provide different advice than DAD does on this subject, or no advice at all in some cases.

The goal of the first blog posting in this series is to address several potential misconceptions that appeared in the email.  I want to start here so as to lay a sensible foundation for the follow-on postings.

Partial Misconception #1: Agile can be prefixed in iteration 0 by architectural design

I’ve named this a “partial misconception” for a few reasons:

1. Disciplined agile teams do some up-front work.  This is called the Inception Phase in DAD, although other methods may refer to it as iteration/sprint 0, warm up, initiation, or other names.  Up-front work is an explicit part of DAD.
2. Iteration 0 isn’t an accurate term.  Although I have used this term in the past when discussing project initiation, the reality is that the average agile team spends about a month doing project initiation activities whereas the average iteration length is two weeks.  So, Inception really isn’t a proper iteration.
3. Inception is more than just architecture.  Several activities typically occur at this point in time, particularly initial architecture envisioning, initial requirements envisioning, initial release planning, and putting the team together to name a few things.

Chapters 6 through 12 in Disciplined Agile Delivery describe these project initiation activities in detail.  Also, I recently wrote that it requires discipline to keep Inception short.

Partial Misconception #2: On principle, Agile is against “big” anything

This is also a “partial misconception” for several reasons:

1. There is in fact a lot of agile rhetoric against big artifacts.  It’s very easy to find agile writings about the challenges with big requirements up front (BRUF), big modeling up front (BMUF) in general, and detailed up front planning for instance.
2. Disciplined agile is against needless waste, not “big” things.  Many traditional modeling and planning practices prove to be quite wasteful in practice.  A serious cultural challenge that the traditional community has is that they are afraid to throw out the bathwater because they assume that the baby will go with it.  I believe that Disciplined Agile Delivery (DAD), and Agile Modeling before it, make it quite clear that it’s possible to gain the benefit of thinking before doing without taking on the very serious problems around doing too much thinking before doing.  So, have the discipline to keep the thinking “baby” yet discard any needless documentation “bathwater”.
3. In rare situations it’s appropriate to create “big” artifacts.  Disciplined agilists aim for sufficient artifacts, the size of which will depend on the context of the situation that your team finds itself in.  In a recent article for Dr. Dobb’s Journal, Disciplined Agile Architecture, I explicitly explored how initial architecture envisioning on an agile project may result in “big” artifacts in some situations.  These situations are very rare mind you, ignoring cultural imperatives to create big artifacts because some people still haven’t made the jump to a disciplined agile approach, but they do happen.  One of the strengths of the DA toolkit is that it is goal driven, not prescriptive, and explicitly explores the tradeoffs surrounding the amount of detail to capture and when to do so.

Partial Misconception #3: Refactoring system architecture beyond mid-implementation is much more expensive than refactoring components

Once again, this is a partial misconception.  I suspect part of the problem is a lack of understanding of what refactoring is really all about, a recurring problem with experienced traditionalists, and part because of a lack of understanding of how architecture is address by disciplined agile teams.  Some thoughts:

1. Refactorings are simple, not difficult.  The goal of refactoring is to make SMALL changes to your design that improve the quality without changing the semantics of the design in a practical manner.  This is true of code refactorings, database refactorings, user interface refactorings, and other types of refactorings.  Small changes are inexpensive to make given the appropriate skills, tools, and organizational environment.
2. Architectural rework (not refactoring) is often difficult.  Rework, or rewrites, are very large changes the goal of which is typically to replace large portions of your solution.  Yes, the later in the lifecycle such rework occurs very likely the more expensive it will be because you’ve built more based on that architecture that is now being reworked.  This is a general issue, not just an agile one.
3. Disciplined agile teams get going in the right direction to begin with. The practice of initial architecture envisioning, which we describe in detail in Chapter 9 of Disciplined Agile Delivery, aims to think through the architectural strategy before getting into construction.
4. Disciplined agile teams prove their architecture works early. The first construction milestone, prove the architecture, reduces the risk of architectural rework.  The goal is to prove that the architecture works by building a working end-to-end skeleton of the solution which implements critical/difficult technical requirements.  This is an agile “fail fast” strategy, or as we say in DAD a “succeed early” strategy, that reduces technical risk on your project.   As an aside, including explicit light-weight milestones such as this is one of many agile governance aspects built right into DAD.
5. Disciplined agile teams have an architectural role.  This role is called Architecture Owner and one of the responsibilities of the person in this role is to guide the team in architectural issues throughout the entire DAD lifecycle.
6. There are no guarantees.   No matter how smart your approach, there’s still a chance that rework can happen.  For instance, you can be mid-way through a project and the vendor of a major architectural component of your solution decides to withdraw it from the market.  Or the vendor goes out of business.  Or perhaps your firm is taken over by another firm and the new owners decide to inflict, oops I mean bless you with, their architectural strategy.  Stuff happens.  Once again, this is a general issue, not specifically an agile one.
7. Quality decreases the cost of rework.  Disciplined agilists will write high-quality code, with a full regression test suite in place, at all times during Construction.  It’s easier to rework high quality artifacts compared with low quality artifacts, so if you get stuck having to perform rework at least the pain is minimized.  My article Agile Testing and Quality Strategies overviews many techniques.

In short, disciplined agile teams do what they can to avoid architectural rework to begin with by having an explicit architecture owner role who focuses on architectural issues throughout the entire lifecycle, by identifying a viable architectural strategy early in the project, proving that architectural strategy works early in Construction, and producing high-quality artifacts throughout the lifecycle that are easier to rework if needed.   With continuous documentation practices and a focus on producing artifacts which are just sufficient enough for the situation at hand, this proves to be far more effective than traditional strategies that assume you require large up-front investments in “big” artifacts, that rely on validation techniques such as architecture reviews instead of the far more concrete feedback of working code, and that often leave quality strategies to the end of the lifecycle (thereby increasing the cost of any rework).

I plan two follow-on blog postings in this series, one exploring how initial architecture envisioning works and one about how to address initial quality requirements (also called non-functional requirements or quality of service requirements) on disciplined agile projects.  Stay tuned!

At Scott W. Ambler + Associates we offer a one-day workshop entitled Agile Architecture: A Disciplined Approach that you should consider if you’re interested in this topic.  We also offer coaching and mentoring services around agile architecture.

Because Disciplined Agile Delivery (DAD) addresses the full delivery lifecycle it explicitly addresses the effort required to transition your solution into production, or in the case of product teams into the marketplace.  This transition effort may be referred to as release, deployment, or even the “end game”.  For teams relatively new to agile this transition effort is a phase, or if you don’t like the term phase then an iteration which very likely has a different time frame than construction iterations.  However, as teams gain more experience with agile and lean techniques the “transition phase” can be evolved into a “transition activity” with a little bit of discipline.  That is the focus of this blog posting.

The DA tool kit is goal-driven, not prescriptive, and as a result it describes the transition effort in terms of goals:

1. Ensure that the solution is ready to deploy.  To achieve this goal you will need to address several issues, such as planning the transition effort, verifying that the solution meets the stakeholder needs for the current release, and validating that the solution is of sufficient quality.  As with other DAD goals, there are several ways each of these issues can be addressed, each with advantages and disadvantages.  There is no single strategy that’s right for all situations.
2. Ensure that stakeholders are ready to receive the solution.  Similarly, there are several issues you need to consider when addressing this goal, including how you communicate that the release is coming, how to train and educate stakeholders in the new solution, and how to prepare the operations and support environment for the new release (one of the many aspects of a disciplined DevOps strategy built right into DAD).
3. Deploy the solution.  Issues to consider when achieving this goal is identifying what should be released, how you can back out the release if you run into trouble, and how to determine that the release was successful.  The last issue, determining success, isn’t as obvious as you would think.  Does success mean that the solution is up and running in production?  Or does success mean that you have satisfied, or better yet, delighted stakeholders?

It is straightforward to empirically observe that the complexity of your transition effort can vary depending on the context of your situation.  For example, a simple standalone application such as a website can be easily deployed regularly because it effectively involves copying some files to a server (farm).  Organizations in this sort of situation may choose to deploy their software many times a day.  However, as we showed in the DAD book, for non-trivial enterprise deployments the transition effort can be significant.    In fact, the November 2010 Agile State of the Art survey found that the average agile team spent six weeks in their transition efforts, with some respondents indicating spending one day or less and some indicating twelve weeks or more.  There are several reasons why this happens:

1. Transition isn’t agile yet. Many organizations are adopting agile techniques for construction but not for other aspects of the delivery lifecycle.  This is often referred to as “Water-Scrum-Fall”.  As a result these organizations are inflicting traditional, labor intensive transition practices on their agile teams.
2. We’re dealing with more than just software.  It’s often a bit more difficult than simply copying some files onto a production server.  Mainstream agile methods describe the goal of creating “shippable software” at the end of every iteration.   That’s a wonderful, overly simplistic philosophy that doesn’t hit the mark in practice. In DAD we prefer to use the phrase “consumable solution” indicating that shippable does not necessarily mean usable software.  Furthermore, in many situations more needs to be delivered than just software – there may also be hardware upgrades, the business processes surrounding the use of the system may evolve, and even the organization structure of the people using the system may evolve.  In short, disciplined agilists adopt a solutions over software mindset.
3. Transition occurs infrequently.  Some agile project teams adopt a long release cycle, often six months or more, and as a result don’t have sufficient opportunities or motivation to get good at deployment.  Sometimes this is due to the nature of the solution they’re working on but more often than not it’s because of the cost of transition – the greater the effort to transition into production the greater the cost to do so, therefore you need to spend more time building the solution to add sufficient value to justify the cost of transition.
4. Lack of vision.  Many IT organizations don’t believe it’s possible that releasing solutions into production can be evolve from a multi-week phase to a multi-hour, or even multi-minute, activity.  Or they realize it’s possible for others to do so but not them (I find it amazing how many people believe they’re in a special situation or blame others, often senior management, for not being able to adopt fundamental strategies that would improve their approach).

Because of the potential complexities of releasing a solution in most mid-to-large sized organizations the deployment of solutions is carefully controlled.  This is particularly true when the solutions share architectures and have project interdependencies, one of the reasons why DAD promotes the need for enterprise awareness within agile teams.  For many reasons release cycles to your stakeholders are less frequent that you would like because of existing complexities within the environment.  However, the ability to frequently deploy value to your stakeholders is a competitive advantage; therefore you should reduce the release cycle as much as possible.  To do so requires a great degree of discipline in areas such as:

1. Quality practices.  First and foremost, testing and any related fixing should be done as much as possible during construction iterations.  Transition is not a “stabilization phase” nor is it a “test phase”, it’s a deployment effort.  Granted, you should still run your regression test suite(s) one last time as part of your overall deployment effort, with the potential that if tests fail you will need to do some fixing, but this should be a minimal effort.  The implication is that quality practices – including testing, refactoring, reviews, code analysis, and so on – be performed continually throughout construction.
2. Pre-production integration and deployment testing.   These are two of the most challenging types of testing, and sadly they’re often given short-shrift in the agile literature.   The goal of pre-production integration testing, sometimes called production testing, is to test the solution in the best approximation of the production environment as possible.  This can be particularly challenging in larger organizations where hundreds of systems are already running in production and dozens are being worked on at any given time.  To make this testing economical during construction many organizations find that they need to go beyond the “whole team testing” approach with a parallel independent testing team that focuses on complex forms of testing that development teams can struggle with.  Similarly deployment testing can also be a challenge because its goal is to determine whether you can successfully deploy into production.
3. Regular coordination between project teams and with operations and support staff. Throughout construction you will want to ensure that the development, operations, and support teams are all aligned.   One aspect of DAD’s enterprise awareness is to include DevOps strategies, such as treating operations and support staff as key stakeholders that you need to work closely with throughout the lifecycle, in the process itself.  The better the coordination the smoother, and therefore quicker, transition will go.
4. Adoption of continuous deployment practices.  The fundamental idea behind continuous deployment is that the more often you choose to deploy the better at it you will get.  This happens because you have more opportunities to hone your deployment skills and also because you’re motivated to find ways to simplify or automate deployment activities.

I’ve seen teams evolve transition phases of several weeks down to several hours through adoption of the disciplined strategies mentioned above.   A disciplined agile team may start with relatively long Inception, Construction, and Transition phases and over time shrink all three down.   Over time the Inception phase mostly disappears, particularly if you maintain team consistency between releases, and as I’ve argued in this posting the Transition phase shrinks to a very small activity.  When deployment becomes inexpensive it enables you to have shorter construction phases and thus more regular releases – teams go from annual releases to bi-annual, then to quarterly releases, then monthly, weekly, and yes, sometimes even daily.  Your team will need to choose a release cadence that makes sense for you.

These days it is fairly easy to observe multi-billion dollar companies, particularly e-commerce companies but even staid organizations such as financial institutions, deploy on a monthly, weekly, and even daily basis.  If other organizations choose to work this way then why can’t you?