Memory Page Size, or kernel page size, refers to the smallest unit of physical memory used for memory management, known as a granule. Operating systems handle the mapping of physical memory addresses to virtual memory addresses, with a memory page being a contiguous block of memory managed by the Memory Management Unit (MMU).

Arm64 architecture supports multiple kernel page sizes, including 4K, 16K, and 64K, enabling system reboot with the appropriate operating system kernel to utilize these sizes. In contrast, the x86 architecture is limited to 4K page sizes.

For memory-intensive applications such as databases, virtualization, or AI inference, using larger pages can enhance cache hit rates and reduce overhead associated with the Translation Lookaside Buffer (TLB). This optimization improves application performance by increasing the efficiency of L1 and L2 cache operations.

• Understanding memory page sizes on Arm64 https://amperecomputing.com/tuning-guides/understanding-memory-page-sizes-on-arm64
• Diagnosing and Fixing a Page Fault Performance Issue with Arm64 Atomics https://amperecomputing.com/tutorials/fixing-page-fault-performance-issue
• Memcached tuning guide https://amperecomputing.com/tuning-guides/memcached-tuning-guide