public inbox for [email protected]
 help / color / mirror / Atom feed
From: Max Kellermann <[email protected]>
To: Pavel Begunkov <[email protected]>
Cc: [email protected], [email protected], [email protected]
Subject: Re: [PATCH 4/8] io_uring/io-wq: cache work->flags in variable
Date: Wed, 29 Jan 2025 20:11:17 +0100	[thread overview]
Message-ID: <CAKPOu+-Mfx9q79nin7tGi1Rr4qGGY=y-2OhuP80U=7EtRpfBdg@mail.gmail.com> (raw)
In-Reply-To: <[email protected]>

On Wed, Jan 29, 2025 at 7:56 PM Pavel Begunkov <[email protected]> wrote:
> What architecture are you running? I don't get why the reads
> are expensive while it's relaxed and there shouldn't even be
> any contention. It doesn't even need to be atomics, we still
> should be able to convert int back to plain ints.

I measured on an AMD Epyc 9654P.
As you see in my numbers, around 40% of the CPU time was wasted on
spinlock contention. Dozens of io-wq threads are trampling on each
other's feet all the time.
I don't think this is about memory accesses being exceptionally
expensive; it's just about wringing every cycle from the code section
that's under the heavy-contention spinlock.

  reply	other threads:[~2025-01-29 19:11 UTC|newest]

Thread overview: 30+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2025-01-28 13:39 [PATCH 0/8] Various io_uring micro-optimizations (reducing lock contention) Max Kellermann
2025-01-28 13:39 ` [PATCH 1/8] io_uring/io-wq: eliminate redundant io_work_get_acct() calls Max Kellermann
2025-01-28 13:39 ` [PATCH 2/8] io_uring/io-wq: add io_worker.acct pointer Max Kellermann
2025-01-28 13:39 ` [PATCH 3/8] io_uring/io-wq: move worker lists to struct io_wq_acct Max Kellermann
2025-01-28 13:39 ` [PATCH 4/8] io_uring/io-wq: cache work->flags in variable Max Kellermann
2025-01-29 18:57   ` Pavel Begunkov
2025-01-29 19:11     ` Max Kellermann [this message]
2025-01-29 23:41       ` Pavel Begunkov
2025-01-30  5:36         ` Max Kellermann
2025-01-30 14:57           ` Jens Axboe
2025-01-31 14:06             ` Pavel Begunkov
2025-01-30 14:54         ` Jens Axboe
2025-01-28 13:39 ` [PATCH 5/8] io_uring/io-wq: do not use bogus hash value Max Kellermann
2025-01-28 13:39 ` [PATCH 6/8] io_uring/io-wq: pass io_wq to io_get_next_work() Max Kellermann
2025-01-28 13:39 ` [PATCH 7/8] io_uring: cache io_kiocb->flags in variable Max Kellermann
2025-01-29 19:11   ` Pavel Begunkov
2025-01-28 13:39 ` [PATCH 8/8] io_uring: skip redundant poll wakeups Max Kellermann
2025-01-31 13:54   ` Pavel Begunkov
2025-01-31 17:16     ` Max Kellermann
2025-01-31 17:25       ` Pavel Begunkov
2025-01-29 17:18 ` [PATCH 0/8] Various io_uring micro-optimizations (reducing lock contention) Jens Axboe
2025-01-29 17:39   ` Max Kellermann
2025-01-29 17:45     ` Jens Axboe
2025-01-29 18:01       ` Max Kellermann
2025-01-31 16:13         ` Jens Axboe
2025-02-01 15:25           ` Jens Axboe
2025-02-01 15:30             ` Max Kellermann
2025-02-01 15:38               ` Jens Axboe
2025-01-29 19:30     ` Pavel Begunkov
2025-01-29 19:43       ` Max Kellermann

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to='CAKPOu+-Mfx9q79nin7tGi1Rr4qGGY=y-2OhuP80U=7EtRpfBdg@mail.gmail.com' \
    [email protected] \
    [email protected] \
    [email protected] \
    [email protected] \
    [email protected] \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox