Abstract: Multi-Processors System-on-Chip (MPSoCs) have emerged as the enabler technology for new computational paradigms such as Internet-of-Things (IoT) and Machine Learning. Network-on-Chip (NoC) communication paradigm has been adopted in several commercial MPSoCs as an effective solution for mitigating the communication bottleneck. The widespread deployment of such MPSoCs and their utilization in critical and sensitive applications, turns security a key requirement. However, the integration of security into MPSoCs is challenging. The growing complexity and high hyper-connectivity to external networks expose MPSoC to Malware infection and code injection attacks. Isolation of tasks to manage the ever-changing and strict mixed-criticality MPSoC operation is mandatory. Hardware-based firewalls are an effective protection technique to mitigate attacks to MPSoCs. However, the fast reconfiguration of these firewalls impose a huge performance degradation, prohibitive for critical applications. To this end, this paper proposes a lightweight broadcasting mechanism for firewall reconfiguration in NoC-based MPSoC. Our solution supports efficient and secure creation of dynamic security zones in the MPSoC through the communication management while avoiding deadlocks. Results show that our approach decreases the security reconfiguration process by a factor of 7.5 on average when compared to the state of the art approaches, while imposing only 11\% area overhead.

Abstract: To address recent many-core architecture design, HPC applications are exploring hybrid parallel programming, mixing MPI and OpenMP. Among them, very few large scale applications in production today are exploiting asynchronous parallel tasks and asynchronous multithreaded communications to take full advantage of the available concurrency, in particular from dynamic load balancing, network, and memory operations overlapping. In this paper, we present our first results of ML-FMM algorithm implementation using GASPI asynchronous one-sided communications and task-based programming to improve code scalability and performance. On 32 nodes, we show an 83.5% reduction on communication costs over the optimized MPI+OpenMP version.