Red Hat Working To Improve Memory Tiering For QEMU VMs To Save RAM

Red Hat engineers are working through a few possible approaches to enhance memory tiering for QEMU-based virtual machines to help in conserving memory. Ultimately they are after leveraging NVMe SSDs in more scenarios for QEMU VMs to save on RAM usage given today's sky high memory pricing. Red Hat virtualization engineer Peter Xu presented at the KVM Forum 2026 this week in Vancouver, Canada on QEMU memory tiering. The goal is for making QEMU VM jive better with memory tiering so that relatively cheaper NVMe SSDs can be engaged in more situations in place of DRAM while still delivering good performance. Among the avenues being evaluated are a new maccess(0 system call, QEMU full user-space tiering with QEMU being responsible for owning the slow tier, or using existing kernel APIs for migrating hotness. That last approach sounds like the lead contender right now with no kernel changes needed and small QEMU changes. That last approach can be further improved upon with addressing kernel limitations of MADV_PAGEOUT behavior via a new kernel patch. Those wishing to learn more can see the presentation page and the KVM Forum video embedded below.