Sat 17 May 2014
Interested in embedded systems, real-time systems, or how the software on satellites or on the Mars rover works, from someone with many years of experience in this field? Joel has a talk just for you.
Joel from RTEMS came to the VU University in Amsterdam to give a talk. The audience was a big group of computer systems researchers that aren't necessarily experts in the field of embedded or realtime.
It turns out the audience was perfect for it because the talk is a great introduction to embedded and realtime; and also goes in-depth when it comes to the unique challenges in SMP scheduling in that area.
We recorded it
My friends at the office Lionel and Kees did audio and video recordings of it. Joel gave me his slides. I transcribed the whole talk (i.e. subtitles) to compensate for the audio being crummy in places. My friend Goran boosted bits of the audio. And I edited it all together.
- What is embedded?
- What is real-time?
- What is an operating system?
- What is an embedded, real-time, operating system?
- What is RTEMS? Where is it used?
- RTEMS architecture
- How are threads managed and scheduled in RTEMS?
- What changes when we start doing all this in SMP
- Audience questions
My favourite part
Is Joel strutting his stuff in response to a question:
Nobody said these were easy. It's multithreaded, that's why we get the big bucks, right?
Watch it to see it!
Here is the video. Enjoy!
PS About The Subtitles
I transcribed all of the subtitles and used the YouTube interface to automatically do the timing for me. This was hit/miss. The first version matched the first 70% of the video perfectly, but always seemed to lose sync completely near the end - same for when the first 50% or all of the video transcription was complete. The next version had different audio and the sync was one big mess.
So I used the first version, adjusted the timing for the new video (the SRT file with timing info was available for download from YouTube), and then was stuck with how to correct the timing for the remaining 30 minutes or so. I call it 'stuck' because I found the youtube interface for correcting these timings far too clumsy to use in this manner. All I needed was displaying the subtitle, and me correcting the timing by hitting a key every time I wanted a transition to the next! Is that so hard? I would be done in the 30 minutes remaining of the video. It can't be done faster. No of course it's not so hard, but that doesn't mean existing tools have an obligation to anticipate your use case, however easy it may be. I looked around but found no subtitle tool, site or software that had implemented my scenario.
Unfortunately the shortest path from my problem to the solution I imagined was using Perl. I wrote a script to read the SRT file up until the timing was wrong, copying over the entries to the output file directly. From then I made it display each subtitle, and wait for a keystroke for each transition; and rewrite the SRT to use the new timings. Somewhat to my surprise, in actual operation, this went perfectly the first time, start to finish. Except for falling asleep once or twice while watching the video again, it was pretty late. I'm not especially proud of the script but I am a bit proud of solving the problem in a completely optimal way in terms of execution time, with fairly low effort in terms of programming time.