Global Adaptation for Energy Efficiency in Multicore Architectures

Abstract

Today mixed-criticality systems are used in most industrial domains, because of their integration advantages. They are smaller, weigh less and reduce the idle time of the previously dedicated hardware. However, these systems can still be improved. Since their hardware is now used more efficiently it automatically suffers more under the aging effects of the heat created by all the simultaneous computations. The heat fastens the aging process of the hardware and increases failure rates. To prevent this the systems need to be cooled down by additional cooling devices like fans. In turn, these devices introduces new failure sources due to their movable parts. In this paper we propose a chip-wide approach to dynamically manage the system computation and communication to optimize the energy-efficiency. By reducing the energy usage of the system we can reduce the additional hardware as well as the weight of the whole system and prolong the system's lifetime as the available power resource lasts longer. We expand the current usage of tile-based energy management to a system wide scheme by implementing a meta-scheduler. This verifiably monitors the system state and changes the schedule if an optimization can be performed.

Global Adaptation Controlled by an Interactive Consistency Protocol

“…The scheduling decisions were made a priori in the meta-scheduler, where the subset of reachable schedules was generated. The simulation shows that the interactive consistency protocol helps to save energy [1].…”

“…At runtime, the difference between the worst case and the real system state can be exploited for the benefit of the system. Using a meta-scheduler [1], a set of schedules can be computed at design time, between which the system can switch at runtime. A so-called meta-scheduler is an application that computes a directed acyclic graph (DAG) of schedules that can be reached at runtime.…”

Adaptive Time-Triggered Multi-Core Architecture

“…The scheduling algorithm for computing the schedules needs to avoid state explosion. For example, enforcing reconvergence horizons for context events prevents an exponential growth of the number of schedules with increasing numbers of context events [19].…”

“…Without reconvergence, there would be an exponential relationship between the number of events in the context model and the schedule size. With a constant reconvergence horizon, there is a polynomial dependency between the number of events and the size of the MG [19].…”