public inbox for [email protected]
 help / color / mirror / Atom feed
* [POSSIBLE BUG] sendmsg ops returning -ETIME
@ 2021-03-26  3:35 Victor Stewart
  2021-03-26 11:26 ` Victor Stewart
  0 siblings, 1 reply; 2+ messages in thread
From: Victor Stewart @ 2021-03-26  3:35 UTC (permalink / raw)
  To: io-uring

i wrote a quic performance tool to research maximal quic server
performance namely syscalls vs io-uring and quic library vs library
(in an apples to apples scenario).

i'm having some issues with the stability of the io-uring networking
layer... and at least thought my rings were just overflowing, but they
aren't. during most runs, all of a sudden submitted sendmsg ops start
failing and returning -ETIME... even though there are absolutely no
timeouts associated with those operations. and it's always that -ETIME
error delivered once things start to fail.

the only performance test implemented so far is a 1GB transfer over
udp over ipv6 loopback. my working guess at the moment is that the
operation pressure rate is exposing some kind of bug somewhere in
io-ring?

i just posted the project on github. the README is basic but i think
provides enough instructions to run the binaries and see for yourself.

let me know what data i can extract so we can quickly rule this
definitely my bug or definitely an io-uring bug. it's really the
bizzare -ETIME that makes me think it might not be mine.

the code is super brief, check perf.networking.cpp

https://github.com/victorstewart/quicperf

^ permalink raw reply	[flat|nested] 2+ messages in thread

* Re: [POSSIBLE BUG] sendmsg ops returning -ETIME
  2021-03-26  3:35 [POSSIBLE BUG] sendmsg ops returning -ETIME Victor Stewart
@ 2021-03-26 11:26 ` Victor Stewart
  0 siblings, 0 replies; 2+ messages in thread
From: Victor Stewart @ 2021-03-26 11:26 UTC (permalink / raw)
  To: io-uring

On Thu, Mar 25, 2021 at 11:35 PM Victor Stewart <[email protected]> wrote:

> let me know what data i can extract so we can quickly rule this
> definitely my bug or definitely an io-uring bug. it's really the
> bizzare -ETIME that makes me think it might not be mine.
>
> the code is super brief, check perf.networking.cpp
>
> https://github.com/victorstewart/quicperf

haha sorry for the false alarm. within 2 seconds of waking up i
realized that i'm reading past the end of the ring!

^ permalink raw reply	[flat|nested] 2+ messages in thread

end of thread, other threads:[~2021-03-26 11:27 UTC | newest]

Thread overview: 2+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2021-03-26  3:35 [POSSIBLE BUG] sendmsg ops returning -ETIME Victor Stewart
2021-03-26 11:26 ` Victor Stewart

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox