Loading...

Exploiting the Intrinsic Redundancy of Multicore Platforms to Achieve Low-power Fault-tolerance in Embedded Applications

Safari, Sepideh | 2016

649 Viewed
  1. Type of Document: M.Sc. Thesis
  2. Language: Farsi
  3. Document No: 48670 (19)
  4. University: Sharif University of Technology
  5. Department: Computer Engineering
  6. Advisor(s): Ejlali, Alireza
  7. Abstract:
  8. VLSI technology scaling has resulted in the integration of a larger number of cores in a single chip in successive technology nodes, offering a great potential to realize task-level redundancy for reliability enhancement in safety-critical applications. However, since battery technology no longer advances commensurately with integration density, multi-core platforms may have limited utility in battery-powered embedded systems. In this thesis, we propose an energy-budget-aware reliability management (enBudRM) method for multi-core embedded systems featuring hybrid energy source (with renewable and non-renewable energy sources). Our method is composed of two phases. In the offline phase, we only consider battery as the energy source and, according to the available energy-budget and slack time for each execution frame, tasks scheduling and voltage-frequency level are determined such that the tasks timing constraints are met while achieving the given reliability target. To increase the battery lifetime, in the online phase, we exploit released slack time at runtime for further voltage scaling. To compensate for the reliability loss of voltage scaling, we exploit an energy harvester along with the battery to enable executing more task replicas. Our experiments show that our energy budgeting method (the offline phase) compared to other approaches reduces the energy consumption on average by 57% (up to 80%). Also, by using harvester we can achieve on average by 35% (up to 45%) battery energy saving, resulting in a higher battery life
  9. Keywords:
  10. Fault Tolerance ; Hard Real Time Systems ; Embedded Real-Time System ; Power Consumption ; Multi-Core Platforms ; Energy Budget Management

 Digital Object List

 Bookmark

No TOC