Yes, you can develop embedded software using agile methodology

The Software Lifecycle Landscape
With the development of embedded hardware, careful attention is given to the design and creation of highly detailed specifications that can be used to source board components. This is usually followed by a phased delivery set of milestones including: prototype, implementation, test supplier qualification and final release to production. This regimen offers the advantage of ensuring quality and functional requirements being met by the time manufacturing. Taking a waterfall approach to hardware development can minimize any risk of downstream issues once hardware goes into volume production.

Software development is perceived to be completely different. The “soft” part in soft ware implies there is an inherent ability to change along the way. Unfortunately, this can give management a false impression that software projects can easily accommodate change with little to no additional cost or schedule impact. The perception of a software developer is further glamourized in Dilbert cartoons by Scott Adams. [1] A programmer is considered to be a breed apart—working in a world combining art, technical inquisitiveness, social discomfort and sometimes even magic. All you need is creative talent, endless hours, lots of coffee, a PC and a handful of software tools.

As hardware development projects are perceived to complete on a timely basis, software development projects are often late and cost much more than was originally planned. Just ask any project manager in the embedded industry who has worked on both hardware and software projects. Given the choice, many project managers would rather manage hardware projects.

Since both software and hardware projects may be under development at the same time, both sets of teams must collaborate. As hardware developers tend to work in a very structured and organized manner, software engineers tend to perform a lot more trial and error. This “software way of life” can be quite a challenge for management and I’ve seen more than one attempt in my career to convert more-disciplined hardware engineers to become programmers. This rarely works due to different skill sets and a radically different mindset.

Rather than demand that software people conform to hardware design approaches, a better way is to take lessons from enterprise and mobile software computing. We must identify the right approach for software development that marries the way software developers like to work with how work is performed.

Audit Your Software Development Lifecycle
Years back, my Datalight software development team was having a tough time juggling incoming product management requests making project schedules difficult to achieve. At the time, any mention of changing our process methodology to agile (specifically, Scrum) was met with, “no way! That stuff is just a fad.”

I found it was easier to let the team identify what didn’t work at the time rather than force fit some newfangled agile methodology on them. So, I asked the team, “OK, what can we improve with our software development process?” Through rather intense brainstorming, we were able to compile a list of improvements to be made:

  1. Decisions made throughout a project lifecycle appear ad hoc and reactive.
  2. QA gets involved way too late.
  3. By the time the customer sees the product, it isn’t what they originally expected.
  4. Some features being developed aren’t getting completed on time. We usually find this out way too late to course correct.
  5. Handoffs at key milestones are ill-defined and by the time the project is completed, the original specifications and design documents aren’t even close to what was actually implemented. And yet, we never have time to update them.
  6. Requirements are often too abstract with little to no mention of priority or guidance for validation.

Next page >>

Comparing Project Management Approaches
In the software development community, there are two competing schools of thought for managing the process of motivating a team to deliver quality products: waterfall and agile. Figure 1 illustrates the differences between the two extremes with waterfall being more structured and less adaptable, while agile (Scrum and eXtreme Programming shown) is less structured and more adaptable.


Figure 1:
Agile versus traditional waterfall (Source: Datalight)

As mentioned earlier regarding hardware projects, waterfall is a step-by-step, phased delivery approach. Completing one step should advance you to the next step as shown in figure 2.


Figure 2:
A waterfall overview (Source: Datalight)

Figure 2 is not exactly linear—instead the phases overlap between milestones. Developers may start the implementation even before the detailed design document has been signed off. Work may spill over to the next milestone work; hence, the term waterfall is appropriate.

For software development, a sequence of phased milestones makes logical sense but, in practice, this approach is not as intuitive as you might think. According to Agile & Iterative Development , the waterfall method has become unattractive to software developers because: [2]

  • Users aren’t always sure what they want and once they see the work, they want it changed.
  • Details usually come out during implementation so committing to upfront schedules is not possible.
  • Approved specs are rarely accurate by the time a project is completed.
  • Disconnected long series of steps with handoffs are typically subjective.
  • Success seems far, far away and, in practice, schedules aren’t very predictable. This results in the team not seeing any work completion for possibly months. And then the problem begins when QA gets their hands on the code.

To compensate for these issues, the Agile Alliance [3] was formed to focus on time-driven, customer efficiency for the software development industry. The Agile Alliance defined core values from W. Edwards Deming’s research on achieving quality and productivity improvements. [4] According to Deming, work can be more efficiently completed with small cross-functional teams focused on immediate quality results.

Each mini-project takes on work by the team to be completed in fixed-length sprints . At Datalight, I’ve found that two-week sprints work best for major product upgrades and one-week sprints work best for product updates or research projects. Throughout each sprint, the team is focused on planning, doing, studying and assessing (PDSA) activities.

The software community at large has embraced this approach most notably with Scrum shown in figure 3. Datalight has selected Scrum as our agile methodology to use.


Figure 3:
A Scrum process overview (Source: Datalight)

Work is defined in T-shirt-sized estimates (small, medium, and all the way up to extra large) at the beginning of a project during a Scrum planning kickoff meeting. The team agrees on resources needed, sprint duration, features (called product backlogs ) and project vision. And the most important work is taken off the remaining work stack in the first sprint planning meeting. The team breaks the work into tasks and assignments so that by the time the sprint ends, the work should have been assigned, designed, implemented and validated. To mitigate risks, the team meets every day to review what work has been accomplished, what is left to complete, and identify where assistance may be needed. Daily meetings may appear to be overkill, but if a daily 15-minute scrum can alleviate project risks, the time is well worth it.

This drastically different approach assumes close-knit collaboration and effective communication. Unfortunately, software developers aren’t usually very proactive at communicating and that’s where a ScrumMaster steps in. A ScrumMaster must facilitate open discussion, communication and closure. This can take lots of extra time to prepare and communicate to the appropriate stakeholders.

A unique characteristic of Scrumming software development projects isthat a product owner and the customer should be represented at Scrumplanning, sprint planning and sprint reviews. During a sprint review,also called a sprint retrospective, the customer has the opportunity toreview progress and the team can adjust to that feedback in the nextsprint.

In traditional waterfall projects, QA involvement takes place nearthe end of a project lifecycle adding significant risk. Until all of thecritical bugs are found and fixed, the long hours of testing and fixingpushes any schedule out. Instinctively, Scrum projects integrate QAtesters as core development team members from the start. Figure4 demonstrates the impact of involving QA too late in the process. Onwaterfall projects, the risk to deliver accelerates, causing long hoursand schedule slips.


Figure 4:
Comparing schedule risk during the product development lifecycle (Source: Datalight)

In agile (Scrum), the risk of late schedules due to quality issuesshould reduce over time. As developers are writing code, the QA testersare busy designing, creating and running tests. I highly recommendinvesting in test automation to allow code to be tested in nightlybuilds without human intervention. The use of test automation iscritical with Scrum projects because of the need to constantly test. Asyou might expect, a software project is not released until all testspass.

Attacking Waterfall Liabilities with Agile
Adopting agile into our software development lifecycle has mitigated many of the liabilities we experienced years ago.


Table 1:
Addressing typical software development liabilities in an agile way (Source: Datalight)

Does Agile Guarantee Success?
Adopting a process, anyprocess, never results in instant success. I cringe every time I hear afellow VP boast, “Yep, we’re going agile and I bet we reduce expensesand meet schedules as a result.” The transition for us at Datalightwasn’t an overnight success either. It took well over a year to get thehang of it. Here are my top gotchas to avoid:

  • Avoid any project restarts by defining and verifying product architecture changes up front. The danger about agile projects is that if dramatic structural changes are left until the end to implement, iterative development cycles may delay possible infrastructure issues. Always prioritize working on those tough underlying framework components, first.
  • Separate uncertainty from focused delivery. The nature of embedded work sometimes enters unchartered territory. In other words, the implementation required to build specific features may take a lot of trial and error to figure out. Attempting to keep a team focused while investigating implementations is asking for trouble. Instead, why not define a short-term research project with short iterations? This mini-project’s sole mission would be to resolve unknown implementation issues before beginning the actual project.

I can think of one example years back where we were planningperformance improvements to our Reliance Nitro embedded file systemtoolkit. Under Linux, we needed to better integrate into the Linuxcache. We determined there were three different approaches we couldtake. Rather than tie up the entire team, a single engineer worked on athree-week research scrum with one-week sprints to prototype eachapproach. The last approach he tried worked beautifully and the researchproject was closed. The team started the actual project the followingweek.

  • Performing a home-grown methodology. Many software organizations that are unable to commit to agile tend to pick and choose parts of competing methodologies into their own secret sauce. By combining Scrum and waterfall (hence, scrummerfall ) into a kinda-agile approach never works. It eventually confuses the team and avoids the benefits that a true agile-run project enjoys.
  • Not handling people issues immediately. As a software manager, I spend as much time working with people issues as I do dealing with technical and process activities. One common characteristic of software developers is the inability to get to done-ness (100 percent completion of tasks). At the daily standups, the team witnesses first hand if estimated tasks are not being completed. Take that opportunity to work with the folks who can’t seem to complete work on time. That guidance can help motivate the individual to improve their performance. The worst outcome is not attending to the problem proactively. Who suffers? The team does as they have to pick up the slack to keep the project on schedule.
  • Lack of coordination between software and hardware engineers. If the hardware team’s implementation schedules are out of sync with the software team’s agile schedule, one will be waiting for the other. This is especially true with embedded projects. Your job is to ensure both groups have coordinated milestones so that the software supports the hardware (and vice versa).

Datalight in an Agile World
It has been years sinceDatalight transitioned to agile. Scrumming has proved to be a greatmotivator to our team by including QA early, breaking up long projectsinto small subprojects, and involving the team to collectively problemsolve on a daily basis. By adopting a collective way to estimate work,our schedules have become much more predictable.

Some of the techniques we use to identify resources have betterenabled us to reserve time for technical support escalation which ourcustomers and sales organization appreciate. Since customers are moreinvolved along the way, our custom services engagements use Scrum as ameans to frequently communicate among project stakeholders. It is allbuilt into the agile DNA!

References

  1. Adams, Scott. Dilbert . dilbert.com.
  2. Larman, Craig. Agile and Iterative Development: A Manager’s Guide. Boston: Pearson Education, 2004.
  3. Agile Alliance. Manifesto for Agile Software Development. (www.agilealliance.org).
  4. Deming, W. Edwards. Out of the Crisis . Boston: The MIT Press, 2000.

4 thoughts on “Yes, you can develop embedded software using agile methodology

  1. “Thanks Ken. Having experienced both success and failure with Waterfall and Agile, I suggest that an optimum approach is neither, rather it is based on perceiving these as a continuum, then positioning along that continuum depending on the characteristics

    Log in to Reply
  2. “Thank you for your kind comments. If there's one lesson I've learned in leading software teams it is to empower the team with easy to understand process. In my experience, mixing different approaches can lead to confusion among the team who wants a few si

    Log in to Reply
  3. “In this field, you may not be able to copy someoneu2019s solution or a practice directly and this is where we all give up and say it will not work for us, but if you custom fit it based on the core principles of agility you are sure to reap the benefits.

    Log in to Reply
  4. “Agile is really directly applicable to HMI development, but less to back-end software and embedded software. We use agile methodology to develop application software. From these experiences we learn to improve our methodology to develop embedded software

    Log in to Reply

Leave a Reply

This site uses Akismet to reduce spam. Learn how your comment data is processed.