netdev - Re: [PATCH net-next] mlx4: optimize xmit path

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] [day] [month] [year] [list]

Date:	Sun, 28 Sep 2014 22:08:43 -0700
From:	Alexei Starovoitov <ast@...mgrid.com>
To:	Eric Dumazet <eric.dumazet@...il.com>
Cc:	Or Gerlitz <gerlitz.or@...il.com>,
	"David S. Miller" <davem@...emloft.net>,
	Jesper Dangaard Brouer <brouer@...hat.com>,
	Eric Dumazet <edumazet@...gle.com>,
	John Fastabend <john.r.fastabend@...el.com>,
	Linux Netdev List <netdev@...r.kernel.org>,
	Amir Vadai <amirv@...lanox.com>,
	Or Gerlitz <or.gerlitz@...il.com>
Subject: Re: [PATCH net-next] mlx4: optimize xmit path

On Sun, Sep 28, 2014 at 7:22 PM, Eric Dumazet <eric.dumazet@...il.com> wrote:
> On Sun, 2014-09-28 at 13:49 -0700, Alexei Starovoitov wrote:
>
>> I see. So xmit_more=true overrides blueflame=on settings.
>
> Yes, unless Mellanox folks have another way.
>
>> I wonder what is the performance difference bf=on vs bf=off,
>> also whether a burst of N packets via bf is slower than
>> burst via queue+doorbell.
>> Some fun exploration for driver experts :)
>
> Prior situation : bf=on : ~4.5 Mpps
>
> queue + doorbell every 8 packets : ~8 Mpps, up to 10Mpps if tuned
> properly.
>
> Rewritten mlx4 tx path and no burst (bf=on doorbell at every packet) :
> 5.3 Mpps
>
> With the full mlx4 patch and burst = 8 -> 14.9 Mpps
>
> This is on a 40Gb NIC, with 108 bytes packets.
>
> (Using <= 104 bytes packets actually gives lower pps because of
> 'inlining' done by the driver :  8 Mpps for PKTSIZE=40 )
>
> # cat /sys/module/mlx4_en/parameters/inline_thold
> 104

nice. Understood. All makes sense.
I'm not sure whether it's worth to break it down
into small pieces. Imo it's good to go as-is.
'inlining' tuning can come later.
Thanks again!
--
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