Memory management in NUMA multicore systems: trapped between cache contention and interconnect overhead