Institute of Networking and Multimedia, National Taiwan University, Jun-14, 2014

Introduction Challenges Motivations Experimental Setup Computational Overhead Application Performance Conclusion

Power management on next-generation mobile devices will be more crucial than the current devices. Existing solutions that borrow legacy designs directly from conventional operating systems, e.g. Linux, are not suitable for mobile devices. Current Scheduler treats all applications equally, and it allocates the available cores and execution cycles to running applications in a fair manner. This paper introduce the concept of application sensitivity into scheduler and governor designs for mobile systems, with the objective to improve user experience and energy efficiency simultaneously.

Determining the sensitivity of each application. Classified in three categories, namely high, medium and low Based on these, rules of sensitivity inheritance and transitions are defined. How to exploit each applications sensitivity during scheduling and governing. Made difficult b/c governor tends to limit the available CPU resources. Solution: manages and allocates CPU resources in certain proportions to applications with high, medium and low sensitivity Validation of the concept and evaluation of the design They implemented it on commercial Samsung Galaxy S3 smartphone.

Scheduler: supervises thread execution, such as selecting the next thread to be executed, the duration and the core to be used. Governor: manages the CPU’s resources by scaling its frequency up or down and by turning the cores on or off. Although vendors may employ different implementations, a general principle is to allocate available CPU resources fairly to application threads and balance the utilization among cores in order to achieve ideal multitasking.

Each core operating at the highest frequency level can provide 1 mega computing cycles per scheduling period T. Zip compressor comprises 4 threads, each which requires 8 mega computing cycles to complete its execution, while the video player has 1 thread that requires 5mega to complete its and 0.5 meg cycles to ensure smooth video laying.

Thread Sensitivity Determination : The first challenge is how to determine the sensitivity of each application to reflect the user’s perception of its delay. Scheduling and Governing Based on Sensitivity : Another challenge is how to perform scheduling and governing based on thread sensitivity so that both user experience and energy efficiency can be improved.

Foreground Domination: A thread’s sensitivity is highly dependent on the application that the user is currently focusing on. Highly Sensitive Interaction: To avoid an adverse impact on user experience, if the user is interacting with the foreground application, we should provide as many CPU resources as possible to ensure a timely response.

Foreground Applications: medium or high sensitivity Background applications: low sensitivity Non-user applications: medium sensitivity Medium -> low: forground is switched to background

Used three apps, with different characteristics, OI file manager: served as an interactive, foreground app. RockPlayer: served as non-inetractive, forground app. Plays 60-second movie trailer with a resolution of 640x360 at 30fps. FtpCafe: used to download 65.9MB file from the internet to create low- sensitivity threads.

The discrepancy in power consumption when the display is on and off is the result of an computational overheads introduced by the new UCSG algorithm. specifically, in every sampling period, UCSG has to update the sensitivity and estimate the computing cycles of each thread.

They advocate that mobile operating systems should move toward user-centric scheduling and governing. They devised a scheduler and governor that allocate CPU resources to mobile applications according to their sensitivity, and implemented our design in Android. The proposed design is particularly suitable when some applications running in the background generates bursty CPU workloads intermittently.

Finished This one

Personalized adaptive scheduling algorithm for multicore smartphones. Switching between cores and the Usage Patterns

Thank You!