I’d like to propose a talk for the upcoming Bangalore Linux meetup on topic: Accelerating Page Migration via Batched Folio Copy and DMA Offload.
Page migration is underlying mechanism for NUMA balancing, memory tiering, memory compaction, and related technologies, yet today it largely relies on single-threaded, folio-by-folio copying. This leads to substantial overhead on modern systems with deep memory hierarchies, large folios, and frequent bulk migration. For example, copying folios between DRAM NUMA nodes
can account for ~25% of the total migration cost when moving 256MB of data.
In this talk, I'll first give an overview of current page migration implementation and then present recent work to enhance Linux page migration through batching, and offloading folio copy to DMA hardware. I'll outline redesigned batched migration flow that pre-copies folios in batch and exposes a pluggable “migrator” interface, enabling acceleration via DMA engines such as AMD PTDMA and SDXI. Using this infrastructure, I demonstrate up to 6x speedups for large-folio migration. I'll also show how the same machinery can be used for memory compaction on busy system by offloading millions of small-page migrations while freeing CPU cycles for user workloads.
Reference:
https://lore.kernel.org/linux-mm/20260309120725.3...@amd.comI’m a Linux kernel developer at AMD, mostly hacking on memory management and NUMA systems. I’ve previously worked at Qualcomm and Intel Labs, contributed to FreeBSD through two Google Summer of Code projects. I enjoy learning from open-source communities and contributing my small pieces to large systems, and I hold a dual B.Tech-M.Tech degree in Electrical Engineering from IIT Kanpur.