netdev - Re: [PATCH net-next] mlx4: exploit skb->xmit_more to conditionally send doorbell

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <CAADnVQLanPQM5vHS4ci27JA1NnzZZirKcedpy=gRieU57=6tNg@mail.gmail.com>
Date:	Thu, 25 Sep 2014 09:08:54 -0700
From:	Alexei Starovoitov <alexei.starovoitov@...il.com>
To:	Eric Dumazet <eric.dumazet@...il.com>
Cc:	David Miller <davem@...emloft.net>,
	netdev <netdev@...r.kernel.org>, Amir Vadai <amirv@...lanox.com>
Subject: Re: [PATCH net-next] mlx4: exploit skb->xmit_more to conditionally
 send doorbell

On Thu, Sep 25, 2014 at 9:05 AM, Eric Dumazet <eric.dumazet@...il.com> wrote:
> On Thu, 2014-09-25 at 08:53 -0700, Alexei Starovoitov wrote:
>> On Thu, Sep 25, 2014 at 7:17 AM, Eric Dumazet <eric.dumazet@...il.com> wrote:
>> > From: Eric Dumazet <edumazet@...gle.com>
>> >
>> > skb->xmit_more tells us if another skb is coming next.
>> >
>> > We need to send doorbell when : xmit_more is not set,
>> > or txqueue is stopped (preventing next skb to come immediately)
>> >
>> > Tested with a modified pktgen version, I got a 40% increase of
>> > throughput.
>>
>> this is awesome!
>>
>> I've been hacking pktgen as well based on Jesper's earlier patch,
>> but in slightly different way. Sounds like you already have working
>> pktgen with xmit_more. Do you mind sharing it? even rough patch
>> would be great.
>
> The plan was to add a 'pburst x' parameter.
>
> Because not only we want to avoid sending the doorbell, but we could
> also not doing the spinlock for every start_xmit().

exactly!
I also hacked
atomic_inc(&(pkt_dev->skb->users))
into
single atomic_add of N packets for bursting to amortize the cost
of both spin_lock and lock xadd

> We also can enforce a global [p]rate (say 100000 packets per second),
> but still allow pktgen to send burts of x packets to use (and test) this
> xmit_more

that would be nice. I'm not sure how yet.

> Nothing ready yet ;)

can't wait :)
--
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html