VMWare ESX centric, but a general problem for any hypervisor: how do you minimize memory access costs when you have a Non-Uniform Memory Access (NUMA) architecture under your SMP abstraction (AMD for a long time, Intel starting with the Nehalem processors).
