Medium
A buffer first-touched on one socket, consumed on the other
On a dual-socket server (two NUMA nodes, one per socket, connected by an inter-socket interconnect such as Intel UPI or AMD Infinity Fabric), a thread pinned to a core on Node 0 allocates and first-touches a large buffer, which the OS's first-touch policy places in Node 0's local DRAM. A different thread, pinned to a core on Node 1, then reads and writes that same buffer heavily. Software sees one flat, cache-coherent shared address space either way. What's the actual hardware-level consequence?
29 of 47 in Computer Architecture