Priority Assignment and Application Mapping for Many-Cores Using a Limited Migrative Model
Ref: CISTER-TR-140204 Publication Date: 11, Feb, 2014
Priority Assignment and Application Mapping for Many-Cores Using a Limited Migrative ModelRef: CISTER-TR-140204 Publication Date: 11, Feb, 2014
Most of present many-core scheduling and application mapping approaches rely on strict assumptions, e.g. uninterruptible availability of all cores, which significantly limits their scope of application. In this paper we consider a Limited Migrative Model (LMM). LMM is the novel approach in the real-time embedded domain, which gives the possibility to efficiently analyse scenarios that involve occasional core shutdowns for various reasons (e.g. energy and thermal management, load balancing, fault tolerance), thus allowing to embrace the full potential and flexibility of many-core platforms. The contributions of this work are as follows. First, we propose the classification of schedulability guarantees for LMM. Then, we introduce various priority assignment and application mapping techniques and subsequently study their impacts on (i) derived schedulability guarantees for critical and time-constrained applications, and (ii) the performance of best-effort workload. The experiments show that LMM is a valuable approach, even when assuming an uninterruptible availability of all cores, while it exhibits clear benefits in cases where, due to necessity to perform occasional core shutdowns, a higher amount of flexibility is desired.