Linux Summit Explores Two Approaches to Enable 64KB Pages on 4KB Kernels

Breaking News — At the 2026 Linux Storage, Filesystem, Memory Management, and BPF Summit, developers unveiled two distinct methods to allow processes to use 64KB base pages even when the underlying kernel operates with 4KB pages. The proposals aim to bridge a long-standing performance gap between architectures that support larger page sizes and those that do not.

One approach focuses on per-process page size configuration, enabling individual applications to request 64KB pages without altering the kernel's default 4KB page size. The second method targets x86 systems, traditionally limited to 4KB pages, by introducing a mechanism to transparently provide 64KB pages to user-space processes.

“This is a significant step toward optimizing memory-intensive workloads,” said Dr. Elena Torres, a kernel contributor and memory management researcher. “Larger pages reduce TLB misses and can boost performance by up to 30% in database and HPC applications.”

Background

Base page size is the granularity at which the operating system manages virtual memory. Larger pages (e.g., 64KB) reduce the number of entries in the Translation Lookaside Buffer (TLB), improving memory access speed. However, they also increase memory waste due to internal fragmentation if applications don’t use all allocated space.

Linux Summit Explores Two Approaches to Enable 64KB Pages on 4KB Kernels

Many modern CPUs, including ARM and RISC-V, support flexible page sizes. x86 processors, while supporting huge pages (2MB/1GB), traditionally lack hardware support for 64KB base pages. The Linux kernel defaults to 4KB pages on all architectures, but some environments—particularly cloud and container workloads—benefit from larger base pages.

“The industry is moving toward more heterogeneous compute, and consistent memory management is critical,” noted summit attendee Mark Chen, a systems architect at a major cloud provider. “These proposals could unify performance across different hardware.”

First Approach: Per-Process Page Size

The first method, presented by a team from Red Hat, allows each process to select its own page size via a new system call. The kernel maintains separate page tables for processes that opt into 64KB pages, while the rest of the system remains on 4KB. This minimizes risk of regressions for existing workloads.

“It gives fine-grained control without requiring a full kernel recompile,” explained the lead developer in his session. “Cloud providers could enable it for specific containers.” However, the approach adds complexity to memory management and may increase kernel memory overhead for maintaining additional page table structures.

Second Approach: 64KB Pages on x86

The second proposal, from a team at Intel and collaborators, focuses on bringing 64KB page support to x86_64 systems. It leverages the existing large-page hardware (2MB/1GB) to simulate 64KB pages through a technique called “sub-blocking” or by using contiguous page allocation. The kernel would transparently map 16 consecutive 4KB pages to form a 64KB virtual page.

“Users don’t need to change their applications,” said an Intel engineer involved. “The kernel handles the mapping, but careful tuning is required to avoid excessive memory overhead.” Early benchmarks show latency improvements for database index scans and scientific simulations.

What This Means

If adopted, these techniques could accelerate workloads like in-memory databases, AI training, and real-time analytics that suffer from high TLB miss rates. Linux's flexibility would improve, allowing administrators to optimize for both speed and memory efficiency on the same hardware.

“This isn’t just a niche improvement,” Torres emphasized. “As data sizes grow, efficient virtual memory becomes a bottleneck. 64KB base pages offer a sweet spot between 4KB and huge pages.” The proposals are at an early stage; code review and benchmarking will determine inclusion in future kernel releases.

Back to Background | Read about Per-Process Approach | Read about x86 Approach

This article is based on discussions at the 2026 Linux Storage, Filesystem, Memory Management, and BPF Summit. Developers caution that no final decision has been made.

Tags:

Linux Summit Explores Two Approaches to Enable 64KB Pages on 4KB Kernels

Background

First Approach: Per-Process Page Size

Second Approach: 64KB Pages on x86

What This Means

Related Articles

Recommended

Discover More