Intel I/OAT reduces system overhead through interrupt moderation, memory access moderation, parallel compute and data movement operations, and reduced context switching. Memory access and data movement efficiency is increased via pre-fetches, a direct memory
access (DMA) subsystem for offloading data movement and asynchronous copies, TCP segmentation offload to the NIC or LOM, and creating affinities between data flows and specific processor cores. These techniques reduce CPU utilization for network I/O tasks, including protocol computation, and significantly reduce CPU stalls caused by cache misses and cache line bouncing.
access (DMA) subsystem for offloading data movement and asynchronous copies, TCP segmentation offload to the NIC or LOM, and creating affinities between data flows and specific processor cores. These techniques reduce CPU utilization for network I/O tasks, including protocol computation, and significantly reduce CPU stalls caused by cache misses and cache line bouncing.