Re: about latencies

[Eric Dumazet <dada1@cosmosbay.com>]

David Miller a écrit :
> From: Eric Dumazet <dada1@cosmosbay.com>
> Date: Fri, 24 Apr 2009 01:07:06 +0200
> 
>> Brandeburg, Jesse a écrit :
>>> On Thu, 23 Apr 2009, Eric Dumazet wrote:
>>>> We could improve this.
>>>>
>>>> 1) dst_release at xmit time, should save a cache line ping-pong on general case
>>>> 2) sock_wfree() in advance, done at transmit time (generally the thread/cpu doing the send)
>>> how much does the effect socket accounting?  will the app then fill the 
>>> hardware tx ring all the time because there is no application throttling 
>>> due to delayed kfree?
>> tx ring is limited to 256 or 512 or 1024 elements, but yes this might
>> defeat udp mem accounting on sending side, unless using qdiscs...
> 
> I'm pretty sure you really can't do this.  It's been suggested
> countless times in the past.
> 
> The whole point of the socket send buffer limits is to eliminate
> the situation where one socket essentially hogs the TX queue of
> the device.


Yes agreed !

Without splitting sk_sleep and enlarging _again_ "struct sock",
cannot we make sock_def_write_space() smarter ? Avoiding scheduling
as the plague Your Honor :)

Dont we have a bit saying there is a sleeping writer ?

We dirty sk_callback_lock, and read "sk_wmem_alloc" and "sk_sndbuf",
we could first test a flag.



Actual function is : (not a patch, just as reference for convenience)

static void sock_def_write_space(struct sock *sk)
{
        read_lock(&sk->sk_callback_lock);

        /* Do not wake up a writer until he can make "significant"
         * progress.  --DaveM
         */
        if ((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
                if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
                        wake_up_interruptible_sync_poll(sk->sk_sleep, POLLOUT |
                                                POLLWRNORM | POLLWRBAND);

                /* Should agree with poll, otherwise some programs break */
                if (sock_writeable(sk))
                        sk_wake_async(sk, SOCK_WAKE_SPACE, POLL_OUT);
        }

        read_unlock(&sk->sk_callback_lock);
}

Thank you