Consistency-Aware and Predictable Memory Processing for Safety-Critical Out-of-Order Multicores

We introduce an approach that facilitates predictable processing of multiple outstanding memory requests in safety-critical out-of-order multicores. A primary challenge addressed by this work is ensuring that multiple outstanding memory requests maintain a memory consistent model while minimizing a low worst-case latency. Adhering to a memory consistency model is crucial for ensuring the correctness of programs executed on such multicores. Our approach, termed predictable processing of multiple outstanding requests $(\mathsf{PPP}$), Teverages micro-architectural enhancements to tackle this challenge. Experimental results show that $\mathsf{PPP}$ delivers speedups of $2.07 \times, 2.79 \times$, and $3.38 \times$ over serialization for 2,4, and 8 cores, while maintaining the worst-case latency.