Section: New Results

Multi-core platforms

We have used memory access traces to map threads on hierarchical multi-core platforms [13] . We have also used software transactional memory to analyze and trace applications running on multi-core architectures [30] .

An approach based on machine learning was used to map threads on transactional memory applications in [31] .

The impact of CPU and memory affinity on multi-core platforms was investigated in [46] using numerical scientific multi-threaded applications as a typical case study. This resulted in improvement of the performance of parallel systems using a NUMA-aware load balancer [68] .

We have also carried a performance evaluation of WiNoCs for parallel workloads based on collective communications [43] as well as for Infiniband networks [40] .