Distributed direct memory operations having DMA engine with frontend and selected backend according to affinity identified based on selected physical address space

A system comprising:

• at least one physical processor;
• a plurality of physical memory components;
• a plurality of registers for storing affinities;
• and
• a direct memory access (DMA) engine comprising:
• a plurality of backends each configured to interface with at least one of the plurality of physical memory components and each having an affinity stored in at least one of the plurality of registers;
• and
• at least one frontend configured to:
• receive, from the at least one processor, a memory operation command;
• read, from the plurality of registers, at least one affinity for one of the plurality of backends;
• select a backend of the plurality of backends available for the memory operation command based on an affinity of the selected backend corresponding to the memory operation command, wherein the affinity is identified based on a physical address space;
• and
• instruct the selected backend to perform the memory operation command.

A method comprising:

• receiving, by a frontend of a direct memory access (DMA) engine, a memory operation command;
• splitting, by the frontend, the memory operation into a plurality of DMA commands;
• selecting, by the frontend for each of the plurality of DMA commands, a backend of a plurality of backends of the DMA engine available for the DMA command based on an affinity of the selected backend corresponding to the DMA command, wherein the affinity is identified based on a physical address space;
• and
• instructing each selected backend to perform the respective DMA command.