The Curious Complexity of Being Turned On
The imaginary Larmin Corp is designing the next killer product: A mood ring. Okay it's too big to wear around your finger and is more of a wrist device. But it works with 80% accuracy and it's got its own app store and it is expected to be a big hit at CES. There's a snag: unnamed sources are attributing the delay in the product launch to the "On/Off" problem. Larmin Corp denied all rumours and promptly launched lawsuits against the unnamed sources, their children, and pets, and the everyone at the bar that night.
Here's how the device works:
- It is comprised of two parts: The mood detector, and the User Interface (UI)
- The user interface runs all the time (It uses your brain waves for energy)
- The mood detector can be turned on and off. However, interpreting your brain waves is complex business, so it can take several seconds to switch on or off.
How do you turn on and off this system? One way is to ignore the delays and pretend it takes no time to turn on and off. The UI simply freezes up until the task is done.
The problem is sometimes mood detector takes a little longer to turn on, and user's think it's crashed and exhibit extreme anger. Some even start banging the entire device on the desk.
So we don't freeze the UI during the turn-on procedure. But this leads to the following behaviour:
As users get impatient waiting for it to turn on, they keep restarting the procedure. But if you try to turn off the detector while it is turning on, it crashes. The UI team first decides to handle this by adding another layer above the mood detector. If you send a command to it, and the mood detector is busy, it stores it in a queue for later. As soon as the mood detector completes, the layer replays the next queued action.
The problem is the user gets impatient and starts repeatedly hitting the button, and the device eventually gets so many commands queued up that it just sits there, repeatedly turning on and off until the user slams it against the wall and the battery falls out. Also, if mood detector ever turns on while the user is angry, it screws up the detector's calibration. (In version 1, users are instructed to be in a neutral mood when activating the ring).
So the design architects bring out the big guns and propose a "OnOffManager". Instead of using a queue of commands, the OnOffManager remembers the last requested state and uses it.
This works pretty well, except that during the design phase the graphics designer gets fed up with the whole debate and and simply grays out the button with an ajax spinny thing, so that any further clicks are ignored during turn on. The OnOffManager code is left in, because it took six months to design, but it is never exercised. Everyone lives happily ever after.
Wait, scratch that. Shortly before release, someone writes a location aware app which periodically turns on the mood detector and sends its status to Facebook. Another group is working on the highly secretive "mood gestures" app, which turns off the mood detector if the user thinks a certain sequence of moods. It's not long before somebody complains about their mood ring randomly turning on and off all the time. After analysis, we see the following sequence:
It's an easy fix. The On/Off manager is modified to keep a count of every app that wants it on. The mood detector is only turned on when the counter goes from 0 to 1, and off when the counter goes from 1 to 0. All other states are ignored.
Everything is great, until the charismatic C.E.O of Larmin Corp, George Jalopsky is giving a keynote speech. During the speech, his mood ring turns on without him realizing it. The video goes viral. Jalopsky flies back in a huff and holds a meeting of all of software development. "You must fix this problem," he cries, waving his wrists around, projecting bright crimson onto the walls and the faces of the engineers. "When I turn it off, I want it to stay off!"
All development is halted and design committees are formed. Soon, no meeting room at Larmin is available because they are all full of developers talking about the problem. Curious discussions like this one are overheard: "If I turn you on, but George turns you off, are you on or are you off?" This is followed by snickers.
And then someone proposes a solution: The mood ring will have a "soft off" function. You can turn it off, but it will still be allowed to turn on again by third party apps, unless you turn it really off. Provisional patents are quickly filed for the software and the design of the power button itself. The software folks toss around the ideas of what off, and really off mean, and whether it makes sense for the ring to be really turned on *snicker*.
Eventually, they come up with the generalized solution. The On/Off manager shall have two counters. One is for a special class of apps that are designated as "System Apps". There's only one for now -- the on off button. But there could be a plurality in the future. The other counter works as before for third-party apps. If the system counter is 0, the mood detector is off and any other commands are ignored. However, if the system counter is non-zero, then the value of the app counter is used to determine if the mood detector should be on. This is illustrated in the following sequence diagram.
And that, folks, is how something like turning the system on and off can grow in complexity very quickly. Soon, Larmin Corp will add low power modes and the special "BlueMood" peripheral, which transmits the moods to other users, but due to brain wave interference patterns, it only works with the mood detector is off even though the user has buttons for both independently.
Come back next time, to read about how the moods are sent from the detector to the display, in "States of confusion".
Exploring sound with WaveletsHere's a program to create scalograms of sound files.
Minimal usable Ubuntu with one commandIf you install the default "ubuntu-desktop" you also get with it a gigabyte of crap that you will never use. But if you don't install the ubuntu desktop, you get a system with a text-only login: prompt, and it's not clear what to install to get it to a usable state.
I have an irrational need to optimize my Ubuntu installation. I did some investigating and came up with this method, which gives a minimal graphical 1.2 GB install, with gnome, networking, and no applications.