[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <CANn89i+atJ0BjQyP4f53jjBEGRaLhdYL=XZoHhT0LnSDEO0SmA@mail.gmail.com>
Date: Mon, 4 Mar 2024 14:14:28 +0100
From: Eric Dumazet <edumazet@...gle.com>
To: Richard Gobert <richardbgobert@...il.com>
Cc: Paolo Abeni <pabeni@...hat.com>, "David S . Miller" <davem@...emloft.net>,
Jakub Kicinski <kuba@...nel.org>, netdev@...r.kernel.org, eric.dumazet@...il.com
Subject: Re: [PATCH net-next 2/4] net: gro: change skb_gro_network_header()
On Mon, Mar 4, 2024 at 2:04 PM Richard Gobert <richardbgobert@...il.com> wrote:
> Overall looks like a great gain for GRO, less code for handling frag0 :)
>
> Could you please share how to measure a <10% gain in pps in a stable
> manner? While perf top is stable for me when testing CPU-bound tasks,
> netperf pps measurements between 2 physical machines generate ~5-7%
> noise when I try to measure.
The pps are measured without "perf top" running.
"sar -n DEV 5 5" , or other non intrusive monitoring tool.
Quite often, noise comes from NUMA hosts, because the NIC has direct
attachment to one of them.
Powered by blists - more mailing lists