Monday, May 10, 2010

Wiki/Issuetracker versus documents

One of the discussions I often encounter in my attempts to help projects improve their tooling, is whether to stick to classic document based documentation or move to a Wiki based documentation platform, supported by a issue tracker.

Most projects actually use a mix of documents, wiki's and issue trackers. Typically the different documentation (meaning online information in general) tools are used for:
  • Documents (MS Word, OO Writer, Google document): Documentation in general, in particular the documentation defined by the development process or as deliverables to the customer.
  • Spreadsheets (MS Excel, OO Calc, Google Spreadsheet): Used for 'things to do' like task lists, backlogs and actions together with mapping information, like requirements, test, etc.
  • Wiki: Wiki's are primarily used for information close to the application development, like updated infrastructure info, development guidelines, build information, etc.
  • Issuetracker: Primarily used for bugtracking.
This 'rich' ecology usually isn't defined by an analyses of how to create an effective tooling platform, but grows out of an ad-hoc need for written information into a online media. This means the information media is chosen based on the rather narrow criteria of what seems most natural for the producer of the information at the time, or in case of process defined documentation, what seems most natural from a organisational point of view.

The problem here is that this approach usually leads to a fragmented information system with documentation being spread over a bunch of decoupled medias. The result is that a lot of time is use on producing and maintaining documentation, which on the other hand is rarely used because of the difficulty in find relevant information.

Instead I'd like to take at step back and take a look at the general qualities one should look for in a good collaborative documentation setup. One of the first steps here is to limit the amount of different medias used for storing information. In the following I will try to argue that document based systems have a number of critical limits, and a wiki/issuetracker based system is a far superior choice.

Here is my list of the most important priorities for a good project information system.
  1. It should be easy to find information. The value of documentation is directly related to how much it is used. This means that the value of documentation which is difficult to find, is of very little use. This is both the case for specific information I may be searching for, but also information which might be relevant for my work, but I'm unaware that it exists. 
  2. It should be easy to contribute with information. The quality of the information in a project is very dependent on the barrier to getting the information from your mind into a persistent media. The more time and effort you need to invest in writing down a piece of information, the less likely it is that is will be added to the project documentation.
  3. Easy access to change information: The amount of information in a project will rise over time and it will quickly become impossible to keep track of all the documentation in the project. Instead project members should be provided with the ability to track the changes to the documentation, which is much easier to cope with. Another benefit of focusing on documentation changes are that this information reflects the areas currently being worked on and are naturally a good point of focus for project members.
  4. It should be easy to focus on the relevant documentation: This is a crucial component in the effort to be able to work with growing amount of information as the project ages. Here it is important to provide focused views on the areas of the project relevant for a given context. Examples of contexts are: backlog, a given feature, a release, architecture, regression test, a given build, my tasks, etc. All these contexts reflects different crosscutting aspects of a project with overlapping information.
  5. Information processing: A lot of the information found in project can be automatically deferred from other information, which is is much better than having to manually produce the information or ignore it. 
A tooling setup which contains these qualities will provide a solid foundation for a project, where a suitable process framework can be added.

So let's take a look at the ability of the different tools to provide the capabilities listed above.

Find information
Information in both document based systems and Wiki/issuetracker information systems can normally both be located by transversing the hierarchical document structure or by free text searching the information (unless you store the document on a simple shared folder). This is okay if you know precisely what you are looking for. But a lot of the useful information we acquire are interesting stuff we stumble across while looking for other things. This typically happens when you browse a website where the hyperlink nature of the documentation allows you to freely lookup more detailed pages describing related information. As the internet shows, this way of build 'clouds' of interlinked information provides a superior user experience compared to a hierarchy of decoupled documents.

Because wikis and (most) issuetrackers are websites in nature, they can take advantage of the webs possibilities for construction integrated information models, where the user can explorer the full information as he choses. In a document based information system you are much more locked into reading the information top to bottom in the sequence defined by the author of the document.

Contribute with information
Let's turn to the differences when adding information to the documentation. The simplest way of allowing users to modify documentation is to make it possible to edit directly in the text or graphics being viewed. For simple text modification to a document or wikipages this is trivial, but what if you want to change the structure of the information elements, eg. split a document in 2, relocate or rename a element. These types of 'refactorings' are performed all the time in wikis and issues tracker, but are rarely seen in document based systems because of the more static nature of the structure found here. This means that an evolutionary development (agile) of a projects information model comes much more natural in a wiki based system than in a document based.

Change information
Both document based systems like Sharepoint/Google Docs and wiki/issuetrackes provide functionality for view change information and subscripting. The main difference here are primarily that the change information typically takes the form of dynamically generated webpages, so it integrates naturally to wiki and issuetracker content, where it more functions as 'metadata' in a document management system.

Focused information
Here we find one of the main differences between the 2 documentation approaches. Because wikis and issuetracker systems have a nice separation between model, logic and view, all information can be collected together in one big shared model, and a infinite number of specialize view can be generated. In a document on the other hand the author writes the view directly, so the only one static view is available. The data contain in the document is also localized to the document, and can not be used by other documentation elements, so the information needs to be written again if they need to appear in other documents.

Generated information
The advantage of the separation between data, logic and presentation is again clear. Here wikis, issuetracker and other webbased applications are able to draw on the full information model for data, and usual comes with extensive out-of-the-box project information processing capabilities. Spreadsheets have a limited capability for processing the data found in the spreadsheet itself, but more ambitious efforts to manipulate the spreadsheet data can quickly become a development project by itself.

To provide rich views in the different tools including information from all relevant areas of the project model, the tools need to be able to display information from the external models contained in the other applications. This can either be accessing the data via webservices or by embedding graphical components from the external tools (widgets/gadgets). These kinds of integrations are an integral part of Wikis and issuetrackers, but are difficult in documents.

On paper the document based system may perform better than described above. It is possible to link across documents, it is possible to refactor document structures, it is possible to embed advanced functionality into documents etc.. In the real world these features are rarely used, and even rarely used with success, because all these functionalities fall outside of the basic nature of documents. That said, we have seen an effort to integrate many of the advantages of the webbased systems into document management system. The most successful challenger here is Google docs, which takes a big step towards making a collaborative document based informationsystem a reality. Many of the document based documentation short comes remains though.

See Tales from history of documentation for a more historical 'analysis' of the nature of documentation.

Tuesday, October 13, 2009

The feature -> task conversion

The inspiration for this post came from a question to my previous post Kanban vs. JIRA, regarding the problem of how to use JIRA to model Kanbans pull based approach to tasks handling. I will try to describe my view on how to address this problem, which isn't a so much a issue with JIRA, but a fundamental task management challenge.

The root of the problem lies in the conversion of the customer/product owners input requirements/features, into a operational set of tasks which can efficiently be processed by the team, and lead to a system with the functionality needed by the customer. This problem manifests itself very different in a Scrum and Kanban oriented project.

In a Scrum project the activity of processing the features contained in the backlog, into processable tasks isn't really addressed. There is a lot of focus on defining and prioritizing features, which is a very important activity, but this doesn't in itself lead to a effective task pipeline (push oriented) or task pool (pull oriented). This isn't necessarily a problem as this conversion activity can just be added to the development process. It can on the other hand turn into a root cause for a lot of problems such as breaking sprints, inablity to finish tasks satisfatory, inferior quality etc, if ignored. This is exactly the case for most of the Scrum projects (and development projects in general) I have encountered, that is, the tasks used to manage the sprints doesn't efficiently reflect all the software development aspects needed to deliver a finished system.

In Kanban the input to the process is a backlog of directly processable tasks, and Kanban therefore requires some kind of mechanism for 'feeding' this task pool. The major benefit of Kanbans approach to task processing is that the requirement of only working with well defined, operational tasks become glaring obvious, and a failure to achieve this will quickly break the Kanban process, because of the teams inablitity to finish tasks. The problem of ensuring that the operational tasks used in the daily work reflects the full set of work need to complete the required system remains, and in must be addressed through activities outside of the Kanban process.

A very good tool in helping with the development and management of the task backlog/pool is defining a number of attributes and task relations which can be use to qualify tasks before thay are added to the 'processable' task pool. These task attributes can also be used by the team 'pulling' the tasks from the backlog to determine which tasks would be the best to work on at a given point in time. The problem is that the task model can become quite complexed as the number of attributes and tasks grows and a good issuetracker, like JIRA, is therefore very useful in managing the model, ensuring consistency and allowing focused views on the aspects of the task model need in a given situation.

Note that I'm not talking about a simple project management tool here, as the tool needs to  include all aspects of the feature -> task lifecycle. Because of this, I will referrer to an issue pool which is the full set of features, bugs, actions, tasks, etc. The task pool, or backlog is the 'processable' task view we can use when accessing the issue model from a task management role. This is relevant, when we work in a Scrum, Kanban or other task oriented context.

Here is my list (very verbose) list of issue attributes I find helpful in nurturing a good issue model together with the relevant JIRA field used for storing the information
  • Summery: This is the 'human readable' key to the issue. A precise and relatively distinct summery is essential in being able to generate efficient issue and task overviews as found in the Scrum and Kanban oriented cardviews, or different tool filters. The summery should be updated as the scope and content of an issue becomes clearer or changes.
  • Detailed description(Description): The full description of the issue. Will properly start out empty and grow as the content of the issue becomes clearer. The ability to include rich content in the issue description is impotent to avoid the need to distribute the issue information over more than one tool.
  • Importance(Priority): Here I prefer a very course grain scale, like the 5 levels found in JIRA. This is partly to allow the team performing the work to have some freedom in choosing which task to work on next (pull task), partly because other issue attributes will contribute other dimensions to the issue model.
  • Iteration(Fixed version): All projects have some kind of time partitioning, like version, increments, sprints, releases, deployment etc. I prefer to use these iteration containers (together with the importance attribute) for modeling issue priority or severity, instead of a very detailed relative prioritization between all the backlog tasks, as described in Scrum. I think that customers (product owners) find this more coarse grained priority system more natural, both because the detailed priority between 2 different issues isn't necessarily a concern to the customer, but also because the iteration containers directly reflects the very tangible deployment and release milestones, which are the real concerns of the customer. It should be possible to placed issues in multiple versions with different granularity, eg. sprint/minor version/major version.
  • Estimated (remaining) time to implement (original and remaining estimate): Always a nice thing to have on a issue when considering when to start work.
  • Issue relations(Links): A major concern not addressed in an one-dimensional backlog, as found in Scrum, is the dependency of tasks on one another. The definition and maintenance of this issue model aspect is a critical input to the team choosing which task to work on. Without a good issue dependency model, it will be difficult to finish tasks, and critical paths will not be addressed.
  • Software domain(Component): An indication of which part of the system under development the task belongs to should be available (if applicable),  eg. database, GUI, business layer, test, analysis, etc.. This can be used to pull tasks according to the competences of the available team resources. 
  • Business domain(Component): Issues should be qualified according to which business domain aspects they reflect (if applicable). This can be view as a high level categorization of user stories/features/use cases.
  • References(Hyperlinks): As issues have relations to other parts of a projects information, eg. requirement, test, stakeholders, etc, this references should also be added to the issue. A very generic way of achieving this is by hyper links in the description.
  • A 2 level breakdown of the issues(Issue-sub task): This reflects the two faced nature of issues. The first (and top level) part is the initial feature oriented aspect of the issue, which is also the part interesting to the project stakeholders, like the sponsor, product owner, users, executives, etc. The second part is the task breakdown of the issue, which is the operational tasks the team can work on. Other project management systems operate with a more complicated 3 or 4 level hierarchical task model, but here I again prefer the multi dimensional model, which much are better at modeling the many aspects of the dynamics in a project.
If the issues are broken down and enriched with all the aspects listed above in a efficient issue tracker, a good foundation for a operational task pool should be available. Of course the road to a correct breakdown and growing of tasks isn't trivial, and reflect the teams increasing insight into the nature of the project. But as I mentioned in the beginning, a efficient and rich model is crucial in the development of a efficient task management system.

So how do we pull task from this multi dimensional issue model. Well, a very pragmatic way of doing this is to define a number of views/filters, which can be used as prioritized mini task pools, from which to pull tasks. An example of such a filter setup could be:
  1. First all Critical issues should be handled.
  2. Secondly all analysis task from the next sprint.
  3. All analysis task from the next version.
  4. All test specifications/acceptance criterias for the next sprint
  5. All task contained in this sprint
  6. All major issues.
  7. .........
This could be flavored with with prioritizing implementing full features, selecting task based on the competences of the free resources, task dependencies, etc.

So in conclusion there is no simple answer on how to produce a efficient task queue where there is a trivial answer about which task to work on next. Gaining insight into the multi-dimentional task model is on the other hand a crucial activity, which can lead to the success or failure of a project, depending on the projects ability to implement a working feature -> task mechanism. Here the rich modeling and filtering capabilities of a efficient issue tracker is critical, especially for more complexed projects. Still, this is secondary to the availability of a competent team, who can do the actual analysis and management work needed to do the actual feature -> task processing.

Monday, August 3, 2009

Kanban vs.JIRA

Some time again I wrote a post, SCRUM vs. JIRA, where I tried to reflect on the difference between the SCRUM and the JIRA project model. Here I argued that SCRUM was based on a number of somewhat fragile preconditions, which made it very difficult to complete sprints successfully. The JIRA model on the other hand is a more 'fundamental', methology neutral breakdown of a projects dynamic aspects, and therefore much more robust basis for a project, where higher order project management frameworks like SCRUM, XP, Unified process, etc can be added.

Now let's take a look at Kanban, which has been recently appeared as a software development framework, to address some of the challenges Scrum is facing. Many of the Scrum related concerns mentioned in the motivation for introducing Kanban corresponds to the list of fragile preconditions I have listed in the SCRUM vs. JIRA posting. Just as I tried to argue in my case for using JIRA as the project model fundation, the Kanban approach to task handling is concerned with focusing on the fundamentals of the development proces, where more ambitious proces frameworks might be constructed (See f.ex Scrum-ban). Let's try to run through the list of Kanban focus areas, and compare these to the task aspects found in JIRA.
  • Task processing: In Kanban the central concern is the efficient processing of tasks, that is the pipeline Open -> In progress -> Resolved. JIRA users will recognize this process as the default JIRA workflow, and this is exactly what JIRA basically is, a application for registering and listing issues/tasks as they move though this lifecycle. A lot of higher order concerns may of course also be modelled in JIRA, but this is optional and can grow together with the project methology as the development process matures. In fact in many cases JIRA is introduced just for this purpose, someone in a project/organisation feels a need for a simple tool for registering and listing the things that needs to be handled in a more robust and shareable manor than by using post-its, simple todo tools, mails, etc.
  • Task exposure: Just as in Scrum, the main visible artifact is the whiteboard (which is the original meaning of Kanban by the way) containing the tasks which are currently being processed. This functions as the task model 'altar' we gather around, synchronize our views of the projects and maintain the models together. This corresponds to JIRA collaborative approach to task management, where the project issues are access through a user-friendly website where all project team members can view and contribute to the model. Note: this is in opposition to many of the more conventional task management tools like MS Project, Excel, etc. where the task management is owned and maintained by a Project Manager, with the occasional showing for the rest of the team.
  • Task pulling: One of the main differences between Scrum and Kanban, is that Kanban focuses on the team choosing which tasks should be processed next (pulled), compared to Scrums empathize on the product owner defining (push) which tasks should handled first (backlog prioritizing) and handled in the near future (sprint planning). Note, this isn't necessarily in contradiction with each other, the task pull and push mechanisms can coexist on two different levels in the project, task pulling is a daily activity, where task pushing is a done on a longer term basis. The Kanban push approach is again seen in JIRA's collaborative approach to updating the status of tasks, where project members 'pull' issues from the pool of open tasks. Any partitioning or prioritization of the open issue pool is a higher order concern, and isn't necessary for maintaining a working task model.
  • Minimizing 'In progress' tasks: This isn't really addressed in JIRA, even though JIRA is a very efficient tool for monitoring this, the 'In progress' list is directly accessible from the project portal page. The problem of building a efficient way of resolving tasks in a consistent and predictable manor, is one of the core software development challenges which Kanban centers around. The mechanisms for handling issue resolution must be found in other tools and human work process, but are crucial for achieving any kind of project success, and certainly for any hope of extending the development process with any kind of more advanced project management methologies (like Scrum).
Now let's take a look at the problematic preconditions I listed in the SCRUM vs. JIRA post and see how Kanban performs compared to Scrum.
  • Stable team: The impact on a Kanban project because of a unstable team is much smaller than on a Scrum project, because a Kanban team doesn't need to concern itself with the sprints failing. The number of tasks processing may become slower, but this is not necessary a problem in Kanban, the decreased velocity can be addressed on a continual basis by adjusting the amount of work in progress. The ability to process tasks may suffer from key competences disappearing from the project in a Kanban project, and might cause difficulty in finishing any tasks in a satisfactory manor, so the Kanban flow might break if this is the case.
  • Stable sprint backlog: The stability of the backlog isn't a Kanban concern, because no predictions is made on when chucks of functionality (sprints) are finished.
  • No external dependencies: Kanban task may of course also depend on external issues, but in Kanban it it much easier to mitigate any problems arising from this, because of the lack of sprints, eg. a task can be rotated out of the 'In progress' pipeline when a 'external block' appears without much interruption to the task processing itself.
  • Clear goal: Not relevant, a clear goal is not a Kanban concern as this is a sprint/iteration concept.
  • Small, well estimated tasks: Essential for Kanban, but this is ok as this is a fundamental precondition for building any kind af predictability into a project.
  • Visual results: Not a Kanban concern, even though it might be a good idea in using this development aspect to assess the task resolution.
As it can be seen, the project requirements needed to get Kanban working efficiently are much more basic than for Scrum, and much closer to the core set of issue management concerns handled in JIRA.

The two constraints which still appear in the Kanban list are that a minimum set of team competences are required and the ability to breakdown work into small, correctly estimated tasks are need. This is pretty much the most basic capabilities you can add to the JIRA issue model, and still get added value in terms of project management. This means that a well-functioning Kanban process is a much more achievable goal than aiming for a Scrum based work process, and will make a robust platform for introducing more advanced methologies, like Scrum. If you on the other hand haven't got the basic issue management aspects addressed in Kanban under control, it will be impossible to get the planning and functionality oriented aspects of Scrum to work, and it will be hard to focus on the real problems in this case, because the symptoms in a Scrum based setup will be much more diverse, hiding the real problems causing the process to break down.

As you might have noticed, I really like Kanban's more basic approach to what lies at the core of good task management (and therefore project management), which introduces a better cause-effect mechanism into software development. This hopefully results in a much improved ability to focus on the projects root causes, compare to more complicated process frameworks.

In a JIRA vs. Kanban context, I also see the issue model found in JIRA much more recognizable in Kanban. This means a good understanding of Kanban and its relation to Scrum, could function as a nice bridge to implementing a smooth 'JIRA -> Kanban -> Scrum -> CMMI focus process' transition as the development process matures (inspired by the process complexity scale found on page 8 of Kanban vs. Scrum).

Tuesday, June 16, 2009

The cost and benefits of documentation

One of my first post on this blog was a rattle about what medias to use, when producing and consuming documentation. Recently I have been lucky enough to define what documentation tools should be used on my projects. The choice has of course fallen on a wiki based information system, more specifically Confluence, backed by JIRA for the more dynamic project information, like bugs, tasks, risks, changes ( I provide consulting services for both products ;-).

But the success in this area has made a new challenge visible, which until now has been hiding in the normal swamp of inefficient and unstructured documentation tooling.

One of the first impulses you might get after getting access to easier-to-use information system, could be to start to document everything, but this ignores one of the major shortfalls of documentation, namely producing and maintaining documentation has a cost. Even after the introduction of more efficient tools for accessing and maintaining documentation , the production and retrieval from the documentation system is far more cumbersome, than using simpler human-to-human information exchange.

On the other hand, documentation also has its benefits, but I find the costs and benefits of what documentation to produce when, are seldom handled in more than a intuitive manor. This results in very simple documentation models. Examples are:
  1. More documentation is better: Heavy process methologies like Waterfall models and typical CMMI implementations are primarily build on the assumption that the mechanism for producing better software is driving the implementation by documentation (Documentation Driven Development, DDD). The idea is that more documentation -> more order and structure. You could say that the focuse here was the benefits of documentation.
  2. All documentation is bad: This was seen in the initial XP and agile movements, which were a counter-reaction to the Waterfall model and other heavyweight methologies (DDD). The burden of documentation is the main focus here.
So in order to get a more balanced approach to when to retain a piece of information in form of documentation and when to stick to a direct person-to-person strategy, is here my list of forces to consider when choosing whether to create a piece of documentation, or not:

Benefits of documentation
  • Doesn't change over time: This is one of the primary drives for creating documentation.
  • Person independent: You do not need to have access to a particular person with the right knowledge to retrieve the documentation , you can just look it up.
  • Scalable: In a pure person-to-person approach to information sharing, you'll quickly find a small subset of project members using a great deal of their time explaining project aspects to other project members. This can be alleviated somewhat by first attempting to looking information up in the project documentation, before turning to the project oracle in the particular area of interest.
  • Geographical invariant: If a project team or stakeholders aren't all placed at the same location, the barrier to person-to-person information exchange rises significantly, thereby making documentation based information exchange more attractive. This is of course only the case if the documentation is accessible at all relevant sites, eg. properly Internet based, or at least intranet based.
  • 24/7 accessibility: Documentation based information can be access all the time. So even when project members have different working hours, vacations etc. they will still (in principle) have access to information generated during their absence.
  • Reference information: Where person-to-person information exchange usually varies according to the context it is used in, documentation based information never changes unless somebody actively updates the documentation. This makes documentation a more stable reference platform, than person based information, which has a tendency to vary more depending on who is delivering the information when. The variations in the information consumers interpretation of the information provided is of coarse another matter.
Costs of documentation
  • Needs work: Compared to human memory based information, the production requires a significant amount of work. This isn't just a resource issue, but may also remove the focus from the real problem being solved, turning the effort into documentation production instead. An example is the development af an application design. The real problem being address here is the developing an efficient application design and applying this to the application implementation. But this can with sufficiently complex design guidelines and inefficient tools turn into a struggle to fill out document templates and placed the produced documents in the right configuration managed structure. The result often being outdated design documentation, never really driving, nor reflecting the application development.
  • Outdated: One of the mentioned benefits of documentation based information, was the stability of the documentation over time. This may be a good ting in case of a static project, eg. project where the project information doesn't change over time, but projects are always changing, so the static aspect of documentation is degrades proportional to the dynamics of the contained information. The degrading is either caused by the documentation becoming outdated, or by the effort need to maintain the documentation.
  • Information retrieval: Several of the benefits mentioned above addresses availability of documentation, eg. you can always find the documentation. The problem is that this apparent quality, doesn't translated into the ability of being able to find the information needed. The information retrieval capabilities of documentation are far inferior to a human ability to interpret and answer questions. This means that the apparent qualities of universal documentation retrieval functionality will be degraded severely by the inability to quickly find the relevant information.
  • Reference interpretation: As mentioned under the documentation benefits, the static nature of document based references doesn't ensure consistent interpretation of the reference information. My claim is that it is actually often possible for an compentant human reference responsible to provide a much easier to understand explanation of how apply the reference information to the situation at hand. Eg. the design says the applcation should look like this, but how does this affect this bit of code I'm working on.
Need driven documentation

All in all there isn't any clear answers on what to document, and it is therefore very difficult to define a standard documentation structure (even though a lot of 'standard' process frameworks attempt to do this). Instead you should cultivate a more agile/intelligent approach in your organization as to what documentation should be generated.

One thing you have to keep in mind is that the (manual) production of a document doesn't contribute anything to adding knowledge to a project, the sole purpose is to retain information according to the the benefits mentioned earlier.

One way of generating just the right amount and type of documentation, is to try to avoid writing documentation before you need it. This doesn't necessary mean activity = document artifact, but that the documentation shouldn't be generated until the distribution and statical benefits of documentation based information artifact becomes apparent. A very efficient mechanism for achieving this, is to consider whether a piece of information should be found in the stored documentation, every time a person asks you a question regarding the project. If the information should be found in the documentation, refer the person to the documentation. This can have several outcomes:
  • The relevant documentation is found, everything is great.
  • The relevant documentation is found, but is obsolete. Here you would update the documentation most of the time. If you chose to not update the documentation, you should consider removed it, avoiding wasting another people time reading invalid documentation.
  • The relevant documentation exists, but isn't found. Here you should consider whether the way of finding this bit of documentation, eg. add links to information, improve searchability, switch to more powerful tools.
  • The relevant documentation doesn't exist. Consider whether the documentation should be produced at this time. Remember, on possible answer to this could be that it isn't worth the effort, I'll just answer the question.
If stick to an documentation generation approach like this, will properly only produce a minimum of documentation, but still over time generate the documentation needed to run the project. This is of course only possible in a very iterative project, very the needs of different kinds of documentation occurs on a repetitive bases, instead of deferring the usage of the documentation until right at the end.

Examples of need driven documentation

An example of this is the test specifications, these should be used continuously throughout the project, not just at an big bang acceptance test at the (apparent) final delivery at the project. Therefore test specification documentation should be produced right before or during the implementation of the functionality addressed in the tests. Using tests in this manor is the foundation for Test Driven Development, where you could says the TDD concept is actually a synergy between implementation and test, because the test specification should also be driven by implementation activities.

Another example is the production of architecture and design documentation. In the 'good old days' these where considered essential to a prober application implementation. But with the event of the more pragmatic agile approach to software development, the awareness of a more application development driven architecture and design emerged. The result is that the value of architecture and design documentation is greatly reduced, and the absence of such documentation isn't any longer considered a sure sign of an chaotic application.

Other documentation drivers
Of course you can't just ignore the more conventional drivers for generating documentation. We all start with a idea of what documentation could be useful in a project, the organization the development project is part of, usually also have some input to what should be produced. But my point here is that these inputs should all be seen in the context of the cost and benefits of documentation, and the decisions on what documentation to produce and maintain should should be based primarily on intelligent need driven ad-hoc decisions, as opposed to poorly understood upfront QA or organizational standards requirements/advices.

Friday, January 9, 2009

Efficient and fun software development the open source way

One of my constant sources of inspiration in how to work with software development, is the way successful Open Source (OS) projects structured. One obvious reason for using OS projects as inspiration, is of course the visibility of how the development works here. But other fundamental forces are at work here which makes OS projects very interesting, when considering how to build an efficient and robust platform for software development.

A interesting aspect of OS projects, it that they have to succeed under conditions which we would normally consider crippling in commercial software development. Some of these conditions are:
  • Allocated resources: One of the cornerstone in every project management model is the ability to plan (and somewhat control) which resources are available when. In OS project people do work, when they have the time and interest.
  • Control by management: In normal commercial projects you have the luxury of having roles,, which are dedicated to controlling the project. These are people like project managers, architects, QA, etc, and are backed by the company or organization. In OS projects there are no formal mechanisms for forcing people to do the 'right' thing.
  • Localized team: Most commercial teams are placed in one location, and distributed teams are usually shunned, because these teams are notorious prone to inefficiency. OS projects are usually distributed, both geographically and in working hours.
  • Sales organization: In commercial organizations, the usage of the products developed are helped on the way by a sales organization persuading customers to the virtues of the products. In OS software the sales organization is usually lacking.
Because of the absence of the listed factors in OS project a number of more fundamental qualities are more clearly visible and have to be addressed.
  • Fun: Because of the voluntary nature of participating in OS projects, an OS project needs to be 'fun' to work in, or else it will died a silent dead because nobody contributes. Many of the following qualities are derived from this.
  • Usage driven documentation: Where, the documentation in commercial projects are usually produced based on what non-users think is a good idea, OS documentation is much more driven by Just-In-Time need for documentation (here non-users are people like QA, project leaders, architects, process definers etc, which don't directly use the developed product or source). The result is, that OS documentation is usually much more relevant and updated than commercial documentation. The OS documentation production is also part of the daily work, so the infamous 'task' of documentation is much less pronounced in OS (eg. boring work isn't done i OS projects).
  • Usage by product and web quality: Where a major part of commercial products success depends on the sales organization, OS projects have to depend more on the merits on the products themselves and the quality of the (typically web based) public interfaces to the consumers of the products. The result is that the usability and visibility of the project is essential, eg. documentation, bug/feature tracking, design/source need to be of high quality.
  • Managerless: OS teams very seldom include non-software producing members, and are very efficient role vice, eg. everybody is producing actual software.
  • High level of automation: Because boring tasks aren't performed in OS projects, these are either automated if they are necessary, or left out if the need of the task is difficult to see. Mature, high quality OS projects therefore have automated many of the repetitive, boring tasks done manually in many commercial projects.
  • Self-organizing: Because projects teams in OS projects aren't defined by outside forces the OS teams usually have a much more organic approach to who-does-what. Everybody is in principle project manager, architect, developer, tester etc.
  • Distributed team: OS projects are in nature spread all over the globe, and project members typically work at different times. To handle this OS projects needs to function efficiently without the people ever meetings of talking together.
For a OS project to be a success these factors have to handled efficiently.

Conclusion: Because of the more 'fundamental' nature of OS software development, a good OS development 'model' can be used as a solid foundation for a commercial development process, where higher order methods for improving software development like SCRUM, Unified Process, CMMI, internal processes, etc. can be added. If, on the other hand, the concerns exposed in OS projects aren't handled in projects focused on higher order development models, like SCRUM, UP, etc. it will be very difficult to make the project a success. And even worse, the forces ruining the project won't be understod, making improvement impossible (my blog SCRUM vs. JIRA and following discussion elaborates a bit on this).

Disclaimer: The differences described between commercial and OS projects are of course exaggerated, many commercial project are adopting more agile approached to software development, which have many similarities to OS development. OS projects are on the other hand seeing greater influence from commercial based team setups, where the advantages of more dedicated contributers becomes available.

Sunday, May 11, 2008

Death by architecture

One of the less mentioned dangers of a waterfall or engineer type approache to software development is the focus on the initial activity of defining and implementation the architecture of the system.

Usually the critic of the 'old' plan-drive methologies runs along the lines of inflexibility and inability to cope with the unknowns of real world software development. This issue is often adressed in 'mature' organizations by assembling a group of experts in the discipline of IT system construction (IT architects), and tasking them with defining the high level technical aspect of the system.

One of first task addressed by the architects is usually to define the non-functional requirements, like performance, availability, scalability, maintainability and so on. These input are used to define the architecture of the system. this includes selecting the technologies to use for building the system. The architecture will often include such things as layer models, security models, persistence models, domain model and a lot of other models for crosscutting concerns.

If the company defining the architecture also provides the platforms for deploying the system the architect will also be tempted to include a lot of enabling technologies in the architecture. This will include a potent cocktail of enterprise, high-availability, cluster, SOA...... middleware, platforms and frameworks which would (in theory) solve most of the non-functional challenges.

This all sounds very clever, but this approach can in the real world be devastation to a project progress. The problem is that the initial focus of the development project will be centered on creating the system architecture, instead of producing visible progress.

If you are lucky you will have access to people understanding the technologies making up the architecture. But more often you will a team with a more generic competence profile, and all the middleware and enterprise platform will remain badly understood black boxes, continually impeding the development of the system, while the promise result gained by using ready-made enterprise software never materializes.

During this phase the trust between the customer and the software development team will deteriorate, because the development team is unable to demonstrate any progress and and understand of the problem domain will not increase significantly. The moral of the development team will also go down, because the members are also unable to see, where the project is going.

Many project have never really gotten through this phase, and have died struggling to define and implement the 'right' architect. If you are able to fight yourself out of the architecture desert, the scars on the customer relationship and the team moral will quite deep and make the work ahead more difficult.

So, how should one avoid this?

As always, the solution is to use a more agile approach. This include following the agile best-practices of:
  • Keep it simple: Do not introduce complications needed because of non-functional requirement, until you can't postpone them any more.
  • Make it work, make it right, make it perform: Avoid focusing on concerns, which wouldn't have an effect until later in the project. You should off course not totally ignore later stage challeges, but the development should always be driven by the shortterm gains.
  • Evolutionary development: Grow the system from a simple seed, nursing and growing the application on a continuous basis.

Wednesday, January 30, 2008


In my role as JIRA evangelist, I have been struggling with making JIRA function efficiently in our departments SCRUM centric environment. The challenge here is that the central SCRUM concept of a sprint is very difficult to model in JIRA.

JIRA's way of partitioning the project into versions might at first glance appear as the same thing as sprints, but a deeper difference between SCRUM's and JIRA's view on software development lurks below. Where JIRA's concept of versions focus on the progress and roadmap of the application, the sprints in SCRUM acts as containers for sets of tasks, which the development team can work on. This indicates a fundamental focus in the two approaches to software projects:
  • JIRA models an application focused approach to software development, with a task management aspect.
  • SCRUM models a task based approach to software development, which hopefully reflects some kind of application development.
In an ideal world this would be two sides of the same thing, and is actually one of the premises SCRUM is build on. But in the real world cracks start to appear in the illusion, that SCRUM is a good way of managing application development.

The difficulty of JIRA to model SCRUM is just one of a number of symptoms that SCRUM is a less than perfect tool for application development. Even though pretty much all of our development teams have at this point succumbed to the euphoria of SCRUMs blessings, no one has really been able to consistently finish 'good' sprint. Usually the sprints ends with a lot of lose ends, and a very unclear picture of what was achieved compared to what was planned.

This is most commonly contributed to a number of more or less unjust factors, like external intervention, unclear goals, imprecise estimates etc. In my opinion all these annoying 'factors' are more an indicator of how fragile the SCRUM sprint is , because of all the implicit requirements the sprint has to its environment. These sprint prerequisites include:
  • Stable team: The central task processing engine in SCRUM is the team. If the team isen't 100% dedicated towards completing the sprint, the sprint breaks. Causes could be the allocation of people for external crisis handling or changes to the partial allocation of team members.
  • Stable sprint backlog: The task set to be processed in a sprint is defined by the sprint backlog. If this changes the sprint breaks. Causes could be the introduction of emergency tasks or just general reprioritizing of tasks by higher powers.
  • No external dependencies: The SCRUM name implies that the team gangs up for completing the sprint at hand and without assistance from others. If the team isn't able to complete the task defined in the sprint alone they have to draw on external resources. If the external resources do not act fast the sprint breaks. This is a very likely result of having external dependecies, which will be outside the control of the SCRUM team and probably have other priorities.
  • Clear goal: One of the cornerstones of the sprint is the team being able to share a common goal and everybody works hard at reaching this goal. If the sprint goal is unclear, much of the 'uniqueness' of the sprint concept disappears, and the sprint 'degenerates' into normal iterations or increments. In my experience the set of tasks very seldom sums up to a common goal, most of the time the sprint backlog is defined by the priorities in the product backlog, ad-hoc bugfixes, available resources and other unrelated tasks.
  • Small, well estimated tasks: Even if the list of tasks in the sprint backlog remains stable, the ratio between the task estimates and the actual work used for completing the tasks has to be pretty stable to be able to predict the velocity of the team. If the velocity of the team is unpredictable the sprint breaks. A lot of the tasks needing to be solved in software development are very hard to estimate individualy. Some examples are bugfixes, analysis work, integration tasks, etc.
  • Visual results: One of the valuable outcomes of good sprints should be clear and visible results of the sprint. If this isn't possible the value of the sprint diminishes significantly. But a lot of the work needing to be done in a software project are enablers for the development of the application, and as such doesn't produce any direct result. These types of tasks include: Framework creation, design and architectural work, documentation and specification work, education, etc
All of the points mentioned above are aspects of software development one should strive to improve. They all contribute to better and faster application development. But in the world I live in, very few of the points are stable enough to act as the foundation for a development method. SCRUMS needs every single one to work perfectly to succeed.

The result is that everybody is currently walking around crying SCRUM as the answer to every project management challenge, but nobody is really seeing it work correctly in their project. Any critical approach to improving this situation has been paralyzed by the blind followers of the SCRUM movement, who have hijacked agile development the last couple of years.