Loading...

Learning-based Task Replication for Reliability Improvement in Multicore Embedded Systems

Siyadatzadeh, Roozbeh | 2022

175 Viewed
  1. Type of Document: M.Sc. Thesis
  2. Language: Farsi
  3. Document No: 55349 (19)
  4. University: Sharif University of Technology
  5. Department: Computer Engineering
  6. Advisor(s): Ejlali, Alireza; Ansari, Mohsen
  7. Abstract:
  8. Cyber-Physical Systems, including embedded systems, have become essential components of various applications in modern life. Due to this fact, cyber-physical and embedded systems must be reliable, safe, and meet timing constraints. Task replication is an effective approach to improve reliability and safety, but it may violate the real-time constraints of the executing tasks and aggravate the aging effects of the system due to elevating the on-chip temperature. Moreover, since embedded systems have time, reliability, power, and thermal constraints, the appropriate number of replicas should be determined to handle the extra overheads. Existing works determine the number of replicas at design time based on the worst-case scenario of the system that causes significant overheads. In order to reduce the mentioned overheads and cope with unpredictable events at run-time, machine learning-based techniques can be exploited to dynamically determine the appropriate number of replicas based on the system's actual-case scenario. In this research, we propose a reinforcement learning-based task replication technique (called RL-TIME) for meeting the reliability target, thermal constraint, and real-time constraints at run-time. Experiments demonstrate that RL-TIME is capable of efficiently reducing power consumption by 63% on average, increasing schedulability by 53% on average, and TSP limitations are met 72% more than others on average when compared to state-of-the-art methods. Note that the overhead of the machine learning approach is included in these results.
  9. Keywords:
  10. Reliability ; Machine Learning ; Multicore Embedded System ; Temperature ; Power Consumption ; Tasks Replication

 Digital Object List