Tuesday, April 21, 2009

Multi-Process Mobile Desktop Concept

With the introduction of the Atom 330 processor, dual-core computing was realized in the mobile world. While still keeping the overall power consumption down, these chips add real power and additional capabilities to the system.

In order to take advantage of this new power, software must be designed with multi-threading in mind. Graphical applications that use a GUI toolkit, often perform background work on one thread, and run the Graphics on another thread. Most of the time, multi-threading can be a headache as developers run into race-conditions, deadlocks, and other problems related to multi-threading. Many times multi-threaded apps are discouraged and avoided by some developers.

In addition, when a thread in an application crashes, it usually brings the entire app down with it. For plugin-centric applications like firefox, this can be a huge annoyance to the end-user when a little ad-blocker plugin brings down your entire browsing session.

So how does a developer take advantage of the multiple core processing power and avoid the issues related to multi-threaded development? The answer may be a multi-process design.

Google Chrome wasn't the first to implement a multi-process design, nor will it be the last. Each tab in the Chrome browser is a separate process. Google did this to increase stability and security. And it works. When a tab crashes in Chrome, the tab dies... Chrome keeps on chugging.

Should the same concept apply to mobile computing? I don't see why it shouldn't. The hardware is becoming cheaper, and multi-core chips are here to stay.

A Multi-Threaded Mobile Desktop

To start out with a mobile desktop, one must first understand the mobile environment. Mobile devices usually hand held, or dash-mounted (in the car), with anywhere from a 4"-8" screen. Because the screen is smaller, fonts and other graphical elements need to scale so they are readable by the user. Applications on these usually run full screen.

The desktop in a mobile environment also differs slightly from a desktop on a personal computer. A PC desktop is usually a place where you can put files, and launch applications. It is usually run by the file-manager like nautilus in GNOME and explorer in windows. A mobile desktop is also a place where you can launch applications. But it probably isn't a good place to dump files on because of the lack of screen space. In reality, the mobile desktop is an application that launches other applications. For our multi-process desktop, we will take this idea one step further.

Building a mobile desktop

Writing apps takes time. Writing plugins using a very specific set of tools and interfaces is often much faster. The desktop in our example will act as a server: it will facilitate communication between plugins. Launch them at the users request, and close them when they are no longer needed. Below is a c++ example of a simple plugin interface:

class PluginInterface

virtual ~PluginInterface();

virtual void OnCommandRecieved(CommandClass command) = 0;
virtual void init(DesktopServer ServerInstance) = 0;
virtual void run() = 0;

Why is communication needed? In a mobile environment, resources are scarce. The audio device is an excellent example of a scarce resource. If you are using your mobile GPS to guide you through traffic while listening to music, you don't want your directions to mix with the music. Not only will you not understand where to go, your music enjoyment will be interrupted by a mix of disharmony and chaos. The GPS application/plugin, needs to ask for control of the audio device. It can send a message to the server to "fade" the music volume down to 0, pause, and then fade back in the music when the GPS is done giving it's direction. That sounds much better than Ozzy Ozborne being interrupted by "Turn Left in 30 meters".

By using DBus, a common *nix IPC messaging system, plugins can communicate with each other, and with the server.

Plugins, however, are usually in-process code. How do they fit in a multi-process environment? Simple, they don't. They need a "harness" in order to run as a separate process. In the case of our mobile desktop, a lightweight "core" will be used. The main function of this "core" is to setup communication with the server, load the plugin, and make sure the plugin gets whatever the server sends it. Below is an simple example of our core application:

int main(int argc, char** argv)
if(argc < 2)
printf("usage: core [pluginname]");

string pluginname = string(argv[1]);

PluginInterface *plugin = LoadPlugin(pluginname);

DesktopServer *Server = new Server();




In our desktop, we will have a dock plugin, an app/plugin launcher plugin, an audio device manager plugin and a notification plugin. When any additional plugins or apps are loaded, an icon will appear on the dock so users can switch between that plugin and others. When a plugin wants to use the audio device, it will ask the audio device plugin for control. Likewise when a plugin wants to notify the user of something, like an incomming call, it can send a message to the notification plugin.

Here is an example of a command being sent to the audio device plugin:

void Speak()
string volumelevel = Server.Evaluate("AudioDevice:GetVolumeLevel");

Server.Send("AudioDevice:Fade '0'");

SpeakSomething("Turn Left in 100 Meters");

Server.Send("AudioDevice:Resume '0'");
Server.Send("AudioDevice:Fade '"+volumelevel+"'");


The server then parses the message, grabs the plugin name "AudioDevice" and forwards the message to the Audio Device plugin:

void OnServerMessageReceived(string message)
string plugin_name = GetPluginName(message);
Plugin plugin = GetPlugin(plugin_name);



Below is a visual example of the server core+plugin concept:

When it's all run together, the result is a tightly integrated stable and secure desktop system like the visual below:

Concept Drawbacks

It would be dishonest to present a concept without pointing out the drawbacks, obvious or not. The first drawback is that applications that are used in the system that aren't plugins or do not have plugin-wrappers have a huge disadvantage. These applications won't be able to integrate at the level the plugins will. They likely won't know about the notification plugin, or the audio device plugin. Second, this system forces the developer to learn yet another plugin system, etc.


A multi-process desktop has many advantages. Stability: plugins can't crash the desktop or any other plugin; Security: precautions made by the OS help partition what the plugin can do; Integration: plugins can talk to each other to provide the ultimate user experience; Rapid Development: plugins are faster to develop than entire apps.

Finally, OpenICE, the Open In-Car-Entertainment group plans on implementing the multi-process desktop for use in in-vehicle infotainment systems in the 2.0 release in 2010. Stay tuned...

No comments: