Experimental io-uring support landed in latest actix-files beta.

The latest actix-files beta has received experimental support for Linux's io-uring, enabling truly async filesystem access.

In our early benchmarks, we're seeing 25% improvement in throughput when serving static files and this should have positive effects on resource usage, too. If you're using Actix Web to serve static files from a Linux machine with a recent kernel, this could be free performance win for you; no code changes necessary.

It's a semver-exempt feature for now, though; so don't be alarmed if it breaks without a major version bump. See actix-files changelog →

Early benchmark on a 4-core machine with a 25Mb file:

tokio::fs (sync thread-pool)

$ wrk -c100 -t4 -d10s http://localhost:8080/big-file
Running 10s test @ http://localhost:8080/big-file
  4 threads and 100 connections
  Thread Stats   Avg      Stdev     Max   +/- Stdev
    Latency     1.43s   332.34ms   2.00s    63.51%
    Req/Sec    24.16     17.88   100.00     76.76%
  657 requests in 10.02s, 16.76GB read
  Socket errors: connect 0, read 0, write 0, timeout 13
Requests/sec:     65.59
Transfer/sec:      1.67GB

io-uring (tokio_uring)

$ wrk -c100 -t4 -d10s http://localhost:8080/big-file
Running 10s test @ http://localhost:8080/big-file
  4 threads and 100 connections
  Thread Stats   Avg      Stdev     Max   +/- Stdev
    Latency     1.13s   224.73ms   1.71s    66.79%
    Req/Sec    35.49     29.28   151.00     78.81%
  825 requests in 10.08s, 21.84GB read
Requests/sec:     81.82
Transfer/sec:      2.17GB