K
KITT
Text
NUMA-Aware Virtual Resource Mapping in Disaggregated Systems
In disaggregated data centers that pool CPU and memory from many machines, access latency is non-uniform both within a box (NUMA) and across boxes (remote resources). This concept covers a mapping algorithm that pins virtual cores and migrates memory across the pool, guided by application performance, resource contention, and utilization, to outperform a NUMA-unaware default scheduler.