public inbox for [email protected]
 help / color / mirror / Atom feed
From: Jens Axboe <[email protected]>
To: Hao Xu <[email protected]>, [email protected]
Cc: Pavel Begunkov <[email protected]>, [email protected]
Subject: Re: [PATCH v2 0/5] fast poll multishot mode
Date: Fri, 6 May 2022 21:08:36 -0600	[thread overview]
Message-ID: <[email protected]> (raw)
In-Reply-To: <[email protected]>

On 5/6/22 8:33 PM, Jens Axboe wrote:
> On 5/6/22 5:26 PM, Jens Axboe wrote:
>> On 5/6/22 4:23 PM, Jens Axboe wrote:
>>> On 5/6/22 1:00 AM, Hao Xu wrote:
>>>> Let multishot support multishot mode, currently only add accept as its
>>>> first comsumer.
>>>> theoretical analysis:
>>>>   1) when connections come in fast
>>>>     - singleshot:
>>>>               add accept sqe(userpsace) --> accept inline
>>>>                               ^                 |
>>>>                               |-----------------|
>>>>     - multishot:
>>>>              add accept sqe(userspace) --> accept inline
>>>>                                               ^     |
>>>>                                               |--*--|
>>>>
>>>>     we do accept repeatedly in * place until get EAGAIN
>>>>
>>>>   2) when connections come in at a low pressure
>>>>     similar thing like 1), we reduce a lot of userspace-kernel context
>>>>     switch and useless vfs_poll()
>>>>
>>>>
>>>> tests:
>>>> Did some tests, which goes in this way:
>>>>
>>>>   server    client(multiple)
>>>>   accept    connect
>>>>   read      write
>>>>   write     read
>>>>   close     close
>>>>
>>>> Basically, raise up a number of clients(on same machine with server) to
>>>> connect to the server, and then write some data to it, the server will
>>>> write those data back to the client after it receives them, and then
>>>> close the connection after write return. Then the client will read the
>>>> data and then close the connection. Here I test 10000 clients connect
>>>> one server, data size 128 bytes. And each client has a go routine for
>>>> it, so they come to the server in short time.
>>>> test 20 times before/after this patchset, time spent:(unit cycle, which
>>>> is the return value of clock())
>>>> before:
>>>>   1930136+1940725+1907981+1947601+1923812+1928226+1911087+1905897+1941075
>>>>   +1934374+1906614+1912504+1949110+1908790+1909951+1941672+1969525+1934984
>>>>   +1934226+1914385)/20.0 = 1927633.75
>>>> after:
>>>>   1858905+1917104+1895455+1963963+1892706+1889208+1874175+1904753+1874112
>>>>   +1874985+1882706+1884642+1864694+1906508+1916150+1924250+1869060+1889506
>>>>   +1871324+1940803)/20.0 = 1894750.45
>>>>
>>>> (1927633.75 - 1894750.45) / 1927633.75 = 1.65%
>>>>
>>>>
>>>> A liburing test is here:
>>>> https://github.com/HowHsu/liburing/blob/multishot_accept/test/accept.c
>>>
>>> Wish I had seen that, I wrote my own! But maybe that's good, you tend to
>>> find other issues through that.
>>>
>>> Anyway, works for me in testing, and I can see this being a nice win for
>>> accept intensive workloads. I pushed a bunch of cleanup patches that
>>> should just get folded in. Can you fold them into your patches and
>>> address the other feedback, and post a v3? I pushed the test branch
>>> here:
>>>
>>> https://git.kernel.dk/cgit/linux-block/log/?h=fastpoll-mshot
>>
>> Quick benchmark here, accepting 10k connections:
>>
>> Stock kernel
>> real	0m0.728s
>> user	0m0.009s
>> sys	0m0.192s
>>
>> Patched
>> real	0m0.684s
>> user	0m0.018s
>> sys	0m0.102s
>>
>> Looks like a nice win for a highly synthetic benchmark. Nothing
>> scientific, was just curious.
> 
> One more thought on this - how is it supposed to work with
> accept-direct? One idea would be to make it incrementally increasing.
> But we need a good story for that, if it's exclusive to non-direct
> files, then it's a lot less interesting as the latter is really nice win
> for lots of files. If we can combine the two, even better.

Running some quick testing, on an actual test box (previous numbers were
from a vm on my laptop):

Testing singleshot, normal files
Did 10000 accepts

________________________________________________________
Executed in  216.10 millis    fish           external
   usr time    9.32 millis  150.00 micros    9.17 millis
   sys time  110.06 millis   67.00 micros  109.99 millis

Testing multishot, fixed files
Did 10000 accepts

________________________________________________________
Executed in  189.04 millis    fish           external
   usr time   11.86 millis  159.00 micros   11.71 millis
   sys time   93.71 millis   70.00 micros   93.64 millis

That's about ~19 usec to accept a connection, pretty decent. Using
singleshot and with fixed files, it shaves about ~8% off, ends at around
200msec.

I think we can get away with using fixed files and multishot, attaching
the quick patch I did below to test it. We need something better than
this, otherwise once the space fills up, we'll likely end up with a
sparse space and the naive approach of just incrementing the next slot
won't work at all.

-- 
Jens Axboe


  reply	other threads:[~2022-05-07  3:08 UTC|newest]

Thread overview: 36+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2022-05-06  7:00 [PATCH v2 0/5] fast poll multishot mode Hao Xu
2022-05-06  7:00 ` [PATCH 1/5] io_uring: add IORING_ACCEPT_MULTISHOT for accept Hao Xu
2022-05-06 14:32   ` Jens Axboe
2022-05-07  4:05     ` Hao Xu
2022-05-06  7:00 ` [PATCH 2/5] io_uring: add REQ_F_APOLL_MULTISHOT for requests Hao Xu
2022-05-06  7:01 ` [PATCH 3/5] io_uring: let fast poll support multishot Hao Xu
2022-05-06 17:19   ` Pavel Begunkov
2022-05-06 22:02     ` Jens Axboe
2022-05-07  6:32       ` Hao Xu
2022-05-07  9:26       ` Pavel Begunkov
2022-05-07  7:08     ` Hao Xu
2022-05-07  9:47       ` Pavel Begunkov
2022-05-07 11:06         ` Hao Xu
2022-05-06 18:02   ` kernel test robot
2022-05-06  7:01 ` [PATCH 4/5] io_uring: add a helper for poll clean Hao Xu
2022-05-06 11:04   ` kernel test robot
2022-05-06 12:47   ` kernel test robot
2022-05-06 14:36   ` Jens Axboe
2022-05-07  6:37     ` Hao Xu
2022-05-06 16:22   ` Pavel Begunkov
2022-05-07  6:43     ` Hao Xu
2022-05-07  9:29       ` Pavel Begunkov
2022-05-06  7:01 ` [PATCH 5/5] io_uring: implement multishot mode for accept Hao Xu
2022-05-06 14:42   ` Jens Axboe
2022-05-07  9:13     ` Hao Xu
2022-05-06 20:50   ` Jens Axboe
2022-05-06 21:29     ` Jens Axboe
2022-05-06  7:36 ` [PATCH v2 0/5] fast poll multishot mode Hao Xu
2022-05-06 14:18   ` Jens Axboe
2022-05-06 16:01     ` Pavel Begunkov
2022-05-06 16:03       ` Jens Axboe
2022-05-06 22:23 ` Jens Axboe
2022-05-06 23:26   ` Jens Axboe
2022-05-07  2:33     ` Jens Axboe
2022-05-07  3:08       ` Jens Axboe [this message]
2022-05-07 16:01         ` Hao Xu

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 \
    [email protected] \
    [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