| layout | title | permalink |
|---|---|---|
page |
PhD Thesis |
/phd-thesis/ |
Android is a versatile and attractive environment for developing embedded systems with a set of applications that is growing well beyond the original market segment its designers had envisioned. The medical industry has invested significant resources in exploring Android as a platform for on-demand, personalized medical devices, e.g., handicap assistive devices, heart monitors, glucose analyzers, fall detectors. The robotic community is also developing control or sensing applications on Android-equipped smartphones, including autopilot control, audio-based indoor localization, harmonized sound reproduction and etc.
{% include image-row.html img1="pics/juav.jpg" caption1="Fig 1.1 Java-based Autopilot Control" img2="pics/turbine.jpg" caption2="Fig 1.2 Wind Turbine Monitor" img3="pics/ci.jpg" caption3="Fig 1.3 Cochlear Implant Simulation" main-caption="Fig.1 Targeting Applications" %}
{% include image.html wrapper-class="image-wrapper400" img="pics/arch.jpg" title="RTDroid System Architecture" caption="Fig.4 RTDroid System Architecture" %}
RTDroid presents a real-time extension on Android for real-time applications, which require strict timing guarantees while still leveraging existing functionalities on Android. More specifically, RTDroid provides a real-time software stack separating from the original Android stack and isolates interferences from non-real-time (Non-RT) apps to real-time (RT) apps while still enabling interaction. Thereby, the development of real-time apps can utilize existing functionalities of Android on RTDroid. RTDroid leverages a real-time kernel, a real-time Java virtual machine and redesigned real-time framework.
We partially patched the Android customised Linux kernel with Linux-RT patch, including real-time scheduling policies, synchronisation primitives. We enable priority awareness in task scheduling and locking as well as basic syscalls for user spaces.
We have ported an existing real-time Java virtual machine, FijiVM, on Android-based Linux kernel for the real-time runtime environment. It provides basic real-time Java programming APIs for the development of the RTDroid framework and has an Ahead-Of-Time (AOT) compiler for static compilation which takes source codes of RTDroid and generates a native binary file.
{% include image.html img="pics/app-manifest.jpg" wrapper-class="image-wrapper400" title="RTDroid Application Manifest Code Snippet" caption="Fig.5 RTDroid Application Manifest Code Snippet" %}
RTDroid provides a redesigned real-time framework implementation for application abstraction and communication APIs. It presents our solution as a real-time programming model that extends Android's basic application components for real-time development. Our programming model consists of real-time components for application expressiveness, real-time channels for interaction and real-time manifest schema for component declarations.
To prevent interferences from the runtime memory management, we alter the traditional garbage collection to an on-stack memory management inspired by the scoped memory of RTSJ. The on-stack memory management preserves memory bound for object allocation for each component, which assures a constant cost for memory reclamation and prevents any dangling reference by using runtime assginment checks.
From a developer's point of view, our real-time components extend Android components, and our real-time application manifest is similar to the Android manifest for component declaration and real-time parameter configuration. Thereby, the real-time parameters are used for scheduling validation and model checking.
{% include image.html img="pics/app-validation.jpg" wrapper-class="image-wrapper500" title="RTDroid Static Application Validation Workflow" caption="Fig.6 RTDroid Static Application Validation Workflow" %}
RTDroid integrates its application manifest with Cheddar*, a real-time scheduling analysis tool, for static application validation. We utilize real-time parameters configured in the RTDroid’s application manifest and feed these parameters into Cheddar for feasibility analysis, including schedulability, memory usage and communication channel overflow.
More details can be found at RTDroid website and the full version of my thesis will be available soon.
* For more information about Cheddar.