[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Message-ID: <CAAeCc_mDM+XygbBDLG+8axXAXgsmh2eUnqdGSFjiLek96LeKxA@mail.gmail.com>
Date: Mon, 2 Sep 2024 11:18:00 +0530
From: Bharat Bhushan <bharatb.linux@...il.com>
To: Jakub Kicinski <kuba@...nel.org>
Cc: Bharat Bhushan <bbhushan2@...vell.com>, netdev@...r.kernel.org,
linux-kernel@...r.kernel.org, sgoutham@...vell.com, gakula@...vell.com,
sbhatta@...vell.com, hkelam@...vell.com, davem@...emloft.net,
edumazet@...gle.com, pabeni@...hat.com, jerinj@...vell.com,
lcherian@...vell.com, richardcochran@...il.com
Subject: Re: [net-next PATCH v7 1/8] octeontx2-pf: map skb data as device writeable
On Thu, Aug 29, 2024 at 8:18 PM Jakub Kicinski <kuba@...nel.org> wrote:
>
> On Thu, 29 Aug 2024 11:17:25 +0530 Bharat Bhushan wrote:
> > > How did you test this prior to adding skb_unshare()?
> > > Could you share some performance data with this change?
> >
> > testing using flood ping and iperf with multiple instance,
>
> Makes sense, neither of these will detect corruption of data pages :(
> IIRC iperf just ignores the data, ping doesn't retransmit.
> You gotta beef up your testing...
>
> > I do not see any drop in performance numbers
>
> Well. What's the difference in CPU busy time of v5 vs v7?
> You'll copy all TCP packets, they are (pretty much) all clones.
cpu is 5-8% more busy when skb unshare.
Thanks
-Bharat
Powered by blists - more mailing lists