Archive
Out Of One, Many
A Software Product Line (SPL) can be defined as “a set of software-intensive systems that share a common, managed set of features satisfying the specific needs of a particular market segment or mission and that are developed from a common set of core assets in a prescribed way“.
The keys to developing and exploiting the value of an SPL are
- specifying the interfaces and protocols between app components and infrastructure components
- the granularity of the software components: 10-20K lines of code,
- the product instantiation and test process,
- the disciplined management of changes to the app and infrastructure components.
- managing obsolescence of open source components/libs used in the architecture
- keeping the requirements and design data in synch with the code base
Any others?
Starting Point
Unless you’re an extremely lucky programmer or you work in academia, you’ll spend most of your career maintaining pre-existing, revenue generating software for your company. If your company has multiple products and you don’t choose to stay with one for your whole career, then you’ll also be hopping from one product to another (which is a good thing for broadening your perspective and experience base – so do it).
For your viewing pleasure, I’ve sketched out the structure and provided some code-level metrics on the software-intensive product that I recently started working on. Of course, certain portions of the graphic have been redacted for what I hope are obvious reasons.
How about you? What does your latest project look like?
Visualizing And Reasoning About
I recently read an interview with Grady Booch in which the interviewer asked him what his proudest technical achievement was. Grady stated that it was his involvement in the creation of the Unified Modeling Language (UML). Mr. Booch said it allowed for a standardized way (vs. ad-hoc) of “visualizing and reasoning about software” before, during, and/or after its development.
Product Line Blueprint
Here it is, the blueprint (patent pending) you’ve been waiting for:
Need a little less abstraction? Well, how about this refinement:
Piece of cake, no? It’s easy to “figure out“:
- the number of layers needed in the platform,
- the functionality and connectivity within and between each of the layers in the stack
- the granularity of the peer entities that go into each layer and which separates the layers
- the peer-to-peer communication protocols and dependencies within each layer
- the interfaces provided by, and required by, each layer in the stack
- what your horizontally, integrate-able App component set should be for specific product instantiations
- how much time, how much money, and how many people it will take to stand up the stack
- how many different revenue-generating product variants will initially be needed for economic viability
- how to secure all the approvals needed
- how to manage the inevitable decrease in conceptual integrity and increase in entropy of the product factory stack over time – the maintenance problem
Perhaps easiest of all is the last bullet; the continuous, real-time management of the core asset base IF the product factory stack is actually built and placed into operation. After all, it’s not like trying to herd cats, right?
Please feel free to use this open source (LOL!) product line template to instantiate, build, and exploit your own industry and domain specific product line(s). Enough intellectualizing and “strategizing” about doing it in useless committees, task forces, special councils, tiger teams, and blue ribbon panels. There’s money to be made, joy to be distributed, and toes to be stepped on; so just freakin’ do it.
Is this post still too abstract to be of any use? Let’s release some more helium from our balloon and descend from the sky just a wee bit more so that we can get a glimpse of what is below us. Try out “revision 0” of this blueprint instantiation for a hypothetical producer of radar systems:
Did you notice the increase in tyranny of detail and complexity as we transcended the 3 levels of abstraction in this post? Well, it gets worse if we continue on cuz we don’t yet have enough information, knowledge, or understanding to start cutting code, building, testing, and standing up the stack – not nearly enough. Thus, let’s just stop right here so we can retain a modicum of sanity. D’oh! Too late!
Fellow Tribe Members
Being a somewhat skeptical evaluator of conventional wisdom myself, I always enjoy promoting heretical ideas shared by unknown members of my “tribe“. Doug Rosenberg and Matt Stephens are two such tribe members.
Waaaay back, when the agile process revolution against linear, waterfall process thinking was ignited via the signing of the agile manifesto, the eXtreme Programming (XP) agile process burst onto the scene as the latest overhyped silver bullet in the software “engineering” community. While a religious cult that idolized the infallible XP process was growing exponentially in the wake of its introduction, Doug and Matt hatched “Extreme Programming Refactored: The Case Against XP“. The book was a deliciously caustic critique of the beloved process. Of course, Matt and Doug were showered with scorn and hate by the XP priesthood as soon as the book rolled off the presses.
Well, Doug and Matt are back for their second act with the delightful “Design Driven Testing: Test Smarter, Not Harder“. This time, the duo from hell pokes holes in the revered TDD (Test Driven Design) approach to software design – which yet again triggered the rise of another new religion in the software community; or should I say “commune“.
BD00′s hat goes off to you guys. Keep up the good work! Maybe your next work should be titled “Lowerarchy Design: The Case Against Hierarchy“.
Dream, Mess, Catastrophe
To build high quality, successful, long-lived, “Big” software, you must design it in terms of layers (that’s why the ISO ISO model for network architecture has 7, crisply defined layers). If you don’t leverage the tool of layering (and its close cousin – leveling) in an attempt to manage complexity, then: your baby won’t have much conceptual integrity; you’ll go insane; and you’ll be the unproud owner of a big ball of mud that sucks down maintenance funds like a Dyson and may crumble to pieces at the slightest provocation. D’oh!
The figure below shows a reference model for a layered application. Note that even though we have a neat stack, we can’t tell if we have a winner on our hands.
By adding the inter-layer dependencies to the reference architecture, the true character of our software system will be revealed:
In the “Maintenance Dream“, the inter-layer APIs are crisply defined and empathetically exposed in the form a well documented interfaces, abstractions, and code examples. The programmer(s) of a given layer only have to know what they have to provide to the users above them and what the next layer below lovingly provides to them. Ah, life is good.
Next, shuffle on over to the “Maintenance Mess“. Here, we have crisply defined layers, but the allocation of functionality to the layers has been hosed up ( a violation of the principle of “leveling“) and there’s a beast in the making. Thus, in order for App Layer programmers to be productive, they have to stuff their head with knowledge/understanding of all the sub-layer APIs to get their jobs done. Hopefully, their heads don’t explode and they don’t run for the exits.
Finally, skip on over to the (shhh!) “Maintenance Catastrophe“. Here, we have both a leveling mess and an incoherent set of incomprehensible (to mere mortals) inter-layer APIs. In the worst case: the layers aren’t discernible from one another; it takes “forever” to on-board new project members; it takes forever to fix bugs; it takes forever to add features; and it takes an heroic effort to keep the abomination alive and kicking. Double D’oh!
Forever == Lots Of Cash
In orgs that have only ever created “Maintenance Messes and Catastrophies“, since they’ve never experienced a “Maintenance Dream“, they think that high maintenance costs, busted schedules, and buggy releases are the norm. How do you explain the color green to someone who’s spent his/her whole life immersed in a world of red?
World Class Help
I’m currently transitioning from one software project to another. After two years of working on a product from the ground up, I will be adding enhancements to a legacy system for an existing customer.
The table below shows the software technologies embedded within each of the products. Note that the only common attribute in the table is C++, which, thank god, I’m very proficient at. Since ACE, CORBA, and MFC have big, complicated, “funky” APIs with steep learning curves, it’s a good thing that “training” time is covered in the schedule as required by our people-centric process. 
I’m not too thrilled or motivated at having to spin up and learn ACE and CORBA, which (IMHO) have had their 15 minutes of fame and have faded into history, but hey, all businesses require maintenance of old technologies until product replacement or retirement.
I am, however, delighted to have limited e-access to LinkedIn connection Steve Vinoski. Steve is a world class expert in CORBA know-how who’s co-authored (with Michi Henning) the most popular C++ CORBA programming book on the planet:
Even though Steve has moved on (C++ -> Erlang, CORBA -> REST), he’s been gracious enough to answer some basic beginner CORBA questions from me without requiring a consulting contract Thanks for your generosity Steve!
Human And Automated Controllers
Note: The figures that follow were adapted from Nancy Leveson‘s “Engineering A Safer World“.
In the good ole days, before the integration of fast (but dumbass) computers into controlled-process systems, humans had no choice but to exercise direct control over processes that produced some kind of needed/wanted results. During operation, one or more human controllers would keep the “controlled process” on track via the following monitor-decide-execute cycle:
- monitor the values of key state variables (via gauges, meters, speakers, etc)
- decide what actions, if any, to take to maintain the system in a productive state
- execute those actions (open/close valves, turn cranks, press buttons, flip switches, etc)
As the figure below shows, in order to generate effective control actions, the human controller had to maintain an understanding of the process goals and operation in a mental model stored in his/her head.
With the advent of computers, the complexity of systems that could be, were, and continue to be built has skyrocketed. Because of the rise in the cognitive burden imposed on humans to effectively control these newfangled systems, computers were inserted into the control loop to: (supposedly) reduce cognitive demands on the human controller, increase the speed of taking action, and reduce errors in control judgment.
The figure below shows the insertion of a computer into the control loop. Notice that the human is now one step removed from the value producing process.
Also note that the human overseer must now cognitively maintain two mental models of operation in his/her head: one for the physical process and one for the (supposedly) subservient automated controller:
Assuming that the automated controller unburdens the human controller from many mundane and high speed monitoring/control functions, then the reduction in overall complexity of the human’s mental process model may more than offset the addition of the requirement to maintain and understand the second mental model of how the automated controller works.
Since computers are nothing more than fast idiots with fixed control algorithms designed by fallible human experts (who nonetheless often think they’re infallible in their domain), they can’t issue effective control actions in disturbance situations that were unforeseen during design. Also, due to design flaws in the hardware or software, automated controllers may present an inaccurate picture of the process state, or fail outright while the controlled process keeps merrily chugging along producing results.
To compensate for these potentially dangerous shortfalls, the safest system designs provide backup state monitoring sensors and control actuators that give the human controller the option to override the “fast idiot“. The human controller relies primarily on the interface provided by the computer for monitoring/control, and secondarily on the direct interface couplings to the process.
Well Architected
The UML component diagram below attempts to model the concept and characteristics of a well architected software-intensive system.
What other attributes of well architected/designed systems are there?
Without a sketched out blueprint like this, if you start designing and/or coding and cobbling together components randomly selected out of the blue and with no concept of layering/balancing, the chance that you end up will a well architected system is nil. You may get it “working” and have a nice GUI pasted on top of the beast, but under the covers it will be fragile and costly to extend/maintain/debug – unless a miracle has occurred.
If you or your management don’t know or care or are “allowed” to expose, capture, and continuously iterate on that architecture/design dwelling in your faulty and severely limited mind and memory, then you deserve what you get. Sadly, even though they don’t deserve it, every other stakeholder gets it too: your company, your customers, your fellow downstream maintainers, etc….
Why am I here?
WTF?
Meh!
D'oh!
