Memory-management short topics: page-table sharing and working sets

The working set work focuses on understanding the usage pattern of a workload, per page type and per NUMA node, with granularity in minutes. It should also have low enough overhead that it doesn't impact application performance by much (to be evaluated), and could be turned on for most latency tolerant workloads to aid proactive reclaim.

The parallel in Brendan Gregg's WSS would be idle page tracking, which tracks accesses by setting and checking the PG_idle bit. The benefit here is that MGLRU already does almost all of this during aging, and is less clunky than userspace writing to /sys/kernel/mm/page_idle/bitmap.

With DAMON, you can get a lot more, e.g. a heatmap, but there's additional work and tuning required to make use of it. The working set extensions are really about exposing information MGLRU already has.