DTA

Digital Theses Archive

 

Tesi etd-09062021-154840

Type of thesis
Dottorato
Author
MASCITTI, AGOSTINO
URN
etd-09062021-154840
Title
Energy-aware Scheduling of Real-Time Tasks on ARM big.LITTLE Architectures
Scientific disciplinary sector
INF/01
Course
Istituto di Tecnologie della Comunicazione, dell'Informazione e della Percezione - PH.D. PROGRAMME IN EMERGING DIGITAL TECHNOLOGIES (EDT)
Committee
relatore Prof. CUCINOTTA, TOMMASO
Presidente Prof. LIPARI, GIUSEPPE
Membro Prof. BEHNAM, MORIS
Membro Prof. DI NATALE, MARCO
Membro Prof.ssa LO BELLO, LUCIA
Keywords
  • ARM big.LITTLE
  • EDF admission test
  • Energy-efficiency
  • Heterogeneous multicore processing
  • Real-time scheduling
Exam session start date
01/11/2021;
Availability
completa
Abstract
This thesis presents Big-LITTLE Constant Bandwidth Server<br>(BL-CBS), a dynamic partitioning approach to schedule<br>real-time task sets in an energy-efficient way on multi-core<br>platforms based on the ARM big.LITTLE architecture. BL-CBS<br>is designed as an on-line and adaptive scheduler,<br>supporting &#39;&#39;open&#39;&#39; systems and based on a<br>push/pull architecture that is suitable to be incorporated<br>in the current SCHED_DEADLINE code base in the Linux<br>kernel. It employs a greedy heuristic to dynamically<br>partition the real-time tasks among the big and LITTLE cores<br>aiming to minimize the energy consumption and the migrations<br>imposed on the running tasks. BL-CBS is then combined with<br>the Task Decomposition technique already proposed in the literature<br>to design a methodology to be used with any Directed Acyclic Graph (DAG)<br>task for partitioning the real-time workload in a transparent way.<br>The new approach is validated<br>through the open-source RTSim simulator, which has been extended<br>integrating an energy model of the ODROID-XU3 board, fitting<br>tightly the power consumption profiles for the big and LITTLE cores of the board.<br>An extensive set of simulations has been run with randomly<br>generated real-time task sets,<br>achieving 15% of energy saving in average with respect<br>to the state of the art GRUB-PA when used with sequential<br>tasks and reaching 10% of energy saving in average over all<br>the performed experiments with respect to GRUB-PA<br>when used with DAG tasks.<br><br>When using BL-CBS in a real system, a key problem is the one of<br>admitting real-time tasks only if the heuristic will be able<br>to find on-line a suitable partitioning of all of the<br>the admitted workload, so to provide the expected guarantees.<br>Therefore, the related problem of admitting real-time tasks onto both a<br>symmertric multi-processor (SMP) and an ARM big.LITTLE platform,<br>where a partitioned EDF-based scheduler is used,<br>is also explored. For the SMP case, it is proposed to<br>combine a well-known utilization-based<br>test for the first-fit partitioning strategy, with a simple heuristic based<br>on the number of tasks and exact knowledge of the utilization of the first<br>few biggest tasks, while for the ARM big.LITTLE case the approach is<br>to combine different formulas for both the non-uniform platform (NUMP)<br>and the SMP cases. This results in effective and efficient tests improving the<br>state of the art in terms of admitted tasks, as shown by an<br>extensive evaluation performed on randomly generated task sets.
Files