Peak Power Management to Meet Thermal Design Power in Fault-Tolerant Embedded Systems

Abstract

Multicore platforms provide a great opportunity for implementation of fault-tolerance techniques to achieve high reliability in real-time embedded systems. Passive redundancy is well-suited for multicore platforms and a well-established technique to tolerate transient and permanent faults. However, it incurs significant power overheads, which go wasted in fault-free execution scenarios. Meanwhile, due to the Thermal Design Power (TDP) constraint, in some cases, it is not feasible to simultaneously power on all cores on a multicore platform. Since TDP is the maximum sustainable power that a chip can consume, violating TDP makes some cores automatically restart or significantly reduce their performance to prevent a permanent damage. This may affect timeliness of the system, and hence, designers face a challenge in deciding how to use multicore platforms in real-time embedded systems. In this paper, at first, we study how the use of passive redundancy (especially for Triple Modular redundancy) can violate TDP on multicore platforms. Then, we propose a scheme for scheduling real-time tasks in multicore systems to conquer the peak power problem in NMR systems. This is because in multicore embedded systems an efficient solution for meeting the TDP constraint is reducing the peak power consumption. The proposed scheme tries to remove overlaps of the peak power of concurrently executing tasks to keep the maximum power consumption below the chip TDP. In the proposed scheme, we devised a policy called PPA-LTF to manage peak power consumption. This policy prevents tasks execution that consume higher power according to the tasks’ power traces. Our experimental results show that our scheme provides up to 50 percent (on average by 39 percent) peak power reduction compared to state-of-the-art schemes.