Intel Continues Optimizing Linux Memory Placement For Optane DC Persistent Memory

With a new patch series for the Linux kernel, memory access performance by one measurement can improve by 116% on a dual socket Intel server with Optane DC Persistent Memory.

Intel open-source developer Huang Ying is seeking feedback on a patch series that allows optimized memory placement in memory-tiered systems, principally those with Optane DC Persistent Memory. Due to persistent memory characteristics being different from conventional DRAM, the patch series works to ensure that hot pages are placed on a DRAM node and migrating hot pages that may get placed in a persistent memory node over to DRAM via NUMA migration. Similarly, cold pages can be migrated to the persistent memory and off the DRAM with related patches published by Intel. The patches do automatically determine the threshold for hot pages.

These auto NUMA kernel patches won't benefit servers without persistent memory. But for an Intel dual Xeon server with Optane DC Persistent Memory, these experimental kernel patches provided: "With all above optimization, the score of pmbench memory accessing benchmark with 80:20 read/write ratio and normal access address distribution improves 116%."

More details on this work via this patch series.