• Queues allocated per core via local memory • IOs submitted on optimal queue for CPU • MSIX vectors mapped to NUMA optimized Queue pair ‘All or Nothing’ optimization, each core gets a queue and a vector or driver reverts to singe queue pair