4 ms·

Up to 2.44x faster for network zerocopy send: https://lore.kernel.org/io-uring/cover.1638282789.git.asml.silence@gmail.com/ https://lore.kernel.org/io-uring/cov

Up to 2.44x faster for network zerocopy send: https://lore.kernel.org/io-uring/cover.1638282789.git.asml.silence@gmail.com/ https://lore.kernel.org/io-uring/cover.1638282789.git.asml.s... Up to 1.6x faster than epoll: https://twitter.com/axboe/status/1362271500489793539 https://twitter.com/axboe/status/1362271500489793539 io_uring 11% faster for non-polled IO: https://twitter.com/axboe/status/1465358880502861829/photo/1 https://twitter.com/axboe/status/1465358880502861829/photo/1 And it's not like io_uring is complete. With every kernel release the gap over epoll gets bigger.

errantmind 5y ago
I was unable to replicate these results, and neither can other people. I'll look into it again though and see. https://github.com/axboe/liburing/issues/189 https://github.com/axboe/liburing/issues/189 edit: I'm not sure why I'm unable to reply to the comment below but I don't think anyone is being unnecessarily combative. I linked it twice because it is relevant to both replies to my comment. There is useful information in that thread if you read it end to end. What it does demonstrate is io_uring isn't a clear cut improvement over a much simpler approach. That may change in the future as it is improved though.
- wtallis 5y ago
  I really don't see why you've chosen to link to that github issue multiple times in this thread. The discussion in that issue doesn't convincingly demonstrate anything except that some people are trying to be unnecessarily combative about io_uring. We don't need to be dragging that attitude into HN threads. If you want to talk about some real performance problems with io_uring, find a way to do so without that baggage.
- CoolGuySteve 5y ago
  I found that io_uring socket latency is about 25% higher than epoll in my own benchmarks. io_uring should be faster by design but it currently doesn’t seem to be. Maybe disk io is a different story.

errantmind 5y ago

I was unable to replicate these results, and neither can other people. I'll look into it again though and see. https://github.com/axboe/liburing/issues/189 https://github.com/axboe/liburing/issues/189 edit: I'm not sure why I'm unable to reply to the comment below but I don't think anyone is being unnecessarily combative. I linked it twice because it is relevant to both replies to my comment. There is useful information in that thread if you read it end to end. What it does demonstrate is io_uring isn't a clear cut improvement over a much simpler approach. That may change in the future as it is improved though.

wtallis 5y ago
I really don't see why you've chosen to link to that github issue multiple times in this thread. The discussion in that issue doesn't convincingly demonstrate anything except that some people are trying to be unnecessarily combative about io_uring. We don't need to be dragging that attitude into HN threads. If you want to talk about some real performance problems with io_uring, find a way to do so without that baggage.
CoolGuySteve 5y ago
I found that io_uring socket latency is about 25% higher than epoll in my own benchmarks. io_uring should be faster by design but it currently doesn’t seem to be. Maybe disk io is a different story.