Linux Swap Size, Swappiness & Ext4 Inode Calculator

Virtual memory management in the Linux kernel orchestrates the delicate balance between high-speed physical RAM, secondary swap storage, and the filesystem page cache. Swap space serves two vital operational roles: acting as an emergency overflow buffer during transient memory allocation spikes, and allowing the kernel to evict inactive anonymous memory pages so physical RAM can be dedicated to hot filesystem dentry and page caches. The aggressiveness of the Linux kernel memory reclamation engine is governed by the vm.swappiness sysctl tunable, which accepts values from 0 to 200 (defaulting to 60 on most distributions). Higher values instruct the kernel to swap out anonymous application pages more readily, whereas lower settings such as 1 to 10 prioritize keeping application working sets resident in physical RAM. On latency-sensitive database servers (PostgreSQL, MySQL) and virtualization hypervisors (Proxmox VE, KVM), setting vm.swappiness=1 minimizes swap-induced latency spikes while preserving emergency OOM-killer safety margins. Alongside virtual memory tuning, Linux ext4 filesystem capacity is bounded by the total number of inode metadata structures allocated at format time by mke2fs. By default, ext4 provisions one inode for every 16,384 bytes (16 KiB) of partition capacity. If an application generates millions of tiny files—such as container image layers, PHP session files, or mail queues—the partition can exhaust 100% of its available inodes and reject writes with a No space left on device error even when gigabytes of raw block storage remain free. An equally important companion sysctl parameter is vm.vfs_cache_pressure, which controls the kernel's tendency to reclaim memory used for directory entry (dentry) and inode caches relative to pagecache and swap. At the default value of 100, the kernel reclaims VFS metadata at a baseline rate. Lowering vm.vfs_cache_pressure to 50 on file servers, busy web origins, or container hosts preserves directory traversal metadata in RAM longer, substantially reducing disk seek and IOPS overhead during frequent stat() and path lookup operations.