Skip to content

Limit multishot receive retries

Jens Axboe io-uring stable 1 version New

Get series with b4
b4 am [email protected]

Versions newest first

  1. v1 Latest4 patches 0 repliesNew

Patches in v1 No applies recorded yet

#Subject and filesSizeRepliesApplied
1/4io_uring/poll: move poll execution helpers higher upio_uring/poll.c+20 -200 replies
2/4io_uring/net: un-indent mshot retry path in io_recv_finish()io_uring/net.c+20 -160 replies
3/4io_uring/poll: add requeue return code from poll multishot handlingio_uring/io_uring.h io_uring/poll.c+15 -20 replies
4/4io_uring/net: limit inline multishot retriesio_uring/net.c+20 -30 replies

Reviews trailers per patch

No Reviewed-by, Acked-by or Tested-by trailers for v1 yet.

Read the v1 thread

Cover letter [PATCHSET 0/4] Limit multishot receive retries

Hi,

If we have multiple receive multishots pending and one/several/all of the clients are flooding us with traffic, then we can retry each multishot receive many times as we keep having data available. Some quick testing here with 8 clients over a 10gbit link, I saw one client do more than 32K retries. This causes an imbalance in how we serve traffic like that, with the result being that each client will see different throughput.

1-2 are just prep patches, no functional changes. Patch 3 doesn't do anything by itself, but it enables the fix that is in patch 4. That patch limits retries to 32, which should be enough to not cause any extra overhead, while still allowing other clients to be processed fairly.

 io_uring/io_uring.h |  8 +++++++-
 io_uring/net.c      | 49 ++++++++++++++++++++++++++++++++-------------
 io_uring/poll.c     | 49 ++++++++++++++++++++++++++-------------------
 3 files changed, 70 insertions(+), 36 deletions(-)
Signature · 1 line
Jens Axboe