LabVIEW Break!

As of a month or so ago I was still going strong with the Eudyptula Challenge tasks: building development kernels, hacking on dynamically loadable modules, exploring intricacies of git, and generally learning my way around the internal structures that make Linux tick.

I had just begun to really dig into the whole idea of concurrency and the mechanisms (locking and so forth) and race condition caveats the kernel and its documentation have to offer. Then something came up.

Cutscene with BBQ

Our local National Instruments LabVIEW User Group runs a pretty tight ship. I haven't used LabVIEW itself much (I have messed around with it quite a bit), but the test equipment I've been working with for the past couple of years runs on LabWindows/CVI -- NI's C environment ("C with training wheels", a coworker of mine likes to say). So I'm not especially proficient in LabVIEW, but I do attend the user group meetings: There's usually a fairly interesting demonstration of its features or of some real-world deployment. The whole concept of dataflow programming is interesting to me for a few reasons, too, related in part to my longtime interest in audio processing -- cf. Pure Data, Max, even (going back a few extra years) Turbosynth.

And I'd be lying if I told you the free barbecue they brought into the meetings wasn't a factor.

Last year the user group offered its regular attendees the chance to take NI's CLAD certification exam at no cost. It took a little bit of study, but the CLAD is their entry-level certification. The exam is fairly basic, and I passed without much difficulty.

If you're not familiar with LabVIEW, it's pretty neat stuff. It's a full-featured high-level language, but the development interface is based on the concept of dataflow. That is, instead of writing text-based code, you connect functional blocks together with graphical "wires". It is used extensively in a number of industries, particularly in testing and industrial data-aquisition applications. If you're just a curious amateur the software isn't cheap, but if (like me) you work in an environment where it is available, it's worth checking out.

Would You Like an Exam with That?

This spring they offered a similar opportunity, but for people who aready have the CLAD certification they offered the CLD exam -- their full professional development certification.

Great, I thought -- not a chance I should pass up. It normally costs about $400. But... the CLD is a far more challenging exam. It's geared not only to proficiency, but to facility with the software and its graphical language.

It's a 4-hour practical exam. They shut you in a room, give you a sealed spec for an application of nontrivial complexity (typical textbook state machines -- modelling behavior of systems like ATMs, vending machines, and so forth -- but with a lot of real-world-like functions, constraints, and corner cases). Without any external references or breaks, you have to build a working application.

Four hours may seem like a long time, but it's widely regarded as the most challenging aspect of the CLD exam. To implement the typically specified complexity in the time allotted, you really have to know your way around the software and be able to make quick decisions about how to implement the application's logic.

There's not any wiggle room. You can't take breaks or use your phone. No Internet or reference manual is available (other than LabVIEW's built-in help). When time runs out, you save your code to a thumb drive, seal it in an envelope, and wait several weeks for an anonymous NI engineer to review your code. To fast-forward a bit: I took the exam back in May, and still haven't received my results.

Cram Time

As I mentioned, I haven't really used LabVIEW that much. When I heard that I could take the CLD exam for free, and that I only had a few weeks to prepare... I did a quick deep-dive into some LabVIEW books, familiarizing myself with some of the conventions and best practices for more intricate applications (things like event-driven logic, queued state machines, and producer-consumer frameworks, among others).

I learned a ton, but I would be shocked if I passed. That is really OK -- getting the certification would have been lovely, but the experience of preparing for and taking the exam were my primary goals. I got a whole lot out of it, not only in that little bout of self-training, but the exam itself was pretty eye-opening, too. Not only will I be far better prepared to approach the exam next time it is offered, but it has given me some ideas for things I could do with LabVIEW in the meantime.

Back to the Penguin

Of course, through all of that, my kernel-coding efforts were on hold. Since the CLD exam, I've spent the last few weeks refamiliarizing myself with the Eudyptula Challenge task I was on when I set it down. It took less time than I feared to find my bearings; I had been in the middle of Task 8, and submitted it just the other day. From what I understand, it takes about a week for Little to respond to that one (one of the more complex tasks), so I'm using the downtime to run through kernel-build processes again, as that's a skill that tends to languish when I don't use it regularly, and it's been a while. I have a feeling I'll be using it again soon.

blogroll

social