Direct I/O

Direct I/O (O_DIRECT) bypasses the OS page cache entirely, giving the application full control over buffering and caching. This avoids double-buffering (application buffer + page cache), eliminates page cache pollution from sequential scans, and reduces memory usage dramatically.

In Glommio

glommio is the only Rust runtime with first-class Direct I/O support. Its DmaFile allocates from io-uring's pre-registered buffer pool and returns reference-counted pointers for zero-copy device-to-user transfer. DmaStreamReader implements intelligent read-ahead that dispatches new reads as each buffer is consumed, maintaining constant I/O parallelism — unlike OS read-ahead which issues a large batch then waits.

Benchmarks on Intel Optane: sequential scans of a 53GB file (2.5x RAM) achieved 7.29 GB/s with Direct I/O vs 0.95 GB/s buffered (7.7x). Random reads: 547K IOPS vs 238K IOPS, using 80KB memory vs gigabytes of page cache.

When to use it

Direct I/O is essential for storage engines, databases, and any workload that manages its own caching. It's counterproductive for general file I/O where the OS page cache is beneficial.